fix(qa): score zio and kyo on the QA page, and unblock zio's stryker run

[kryptt]

Rebased onto main (ad30b87). Four PRs merged while this was open, and two
of them changed what this PR is even about. The corrections are called out inline;
the short version is that zio no longer renders n/a, because it no longer has
zero mutants.

The bug

site/tools/gen-qa-report.py builds the mutation table from a hard-coded
MUTATION_MODULES list, and that list stopped at jsoniter. zioIntegration
and kyoIntegration have been in the mutationAll alias for a while, so stryker
has been running over both — but the generator never looked for their reports, so
neither module has ever had a row on
quality-assurance.md. No error, no empty row,
just absence.

The two lists use different identifiers, and that is the trap the bug is made
of: quality.yml's loop takes sbt project ids (zioIntegration,
schemesLaws), MUTATION_MODULES takes on-disk directories (zio,
schemes-laws, which latest_report globs). Getting it backwards silently finds
nothing — exactly the failure being fixed — so both lists now carry a comment
saying so and pointing at each other.

A third instance of the same drift, fixed here: quality.yml's per-module loop
was generics schemes circeIntegration avroIntegration jsoniterIntegration —
missing schemesLaws as well as both integrations. A module absent from that
loop produces no report.json on the release run, so its row falls back to
dashes, or keeps whatever a developer's local run left on disk. The loop now
matches the mutationAll alias.

What the rebase changed

zio no longer scores n/a. When this PR was written, zio/ was a 98-line
DI integration — ZEnvironment / ZLayer / Ref wiring delegating straight into
ZIO's API — and stryker generated zero mutants for it, so n/a was the honest
row. #92 then landed the ecosystem optics (zio-schema DynamicValue kit, zio-json
AST + JsonCursor, STM focus ops, Chunks) and with them a lot of genuinely
mutatable code.

On the rebased tree project zioIntegration; stryker does not print n/a. It
aborts, with the same exception #109 fixed for kyo, now in
zio/json/JsonOptics.scala:

stryker4s.exception.UnableToFixCompilerErrorsException: Unable to remove non-compiling mutants
  'Extension without extension methods'
  'Not found: type To' / 'Not found: self' / 'Not found: type A'

Same mechanism, same fix, so the second commit applies it: stryker4s 0.20.3
re-prints every mutated file through scalameta, and a single-method
significant-indentation extension clause is not a Term.Block, so it comes back
as the one-line form with the method's leading Scaladoc replayed between the two —
the forced newline lands def in column 0. Four such clauses exist in the module:
JsonCursor.optional and JsonCodec.stringPrism in json/JsonOptics.scala,
Schema.dynamicPrism and BinaryCodec.prism in schema/SchemaOptics.scala.
Bracing their bodies makes the printer emit the braced form and the newline
harmless. ZioOptics's single-method clause already hoists its Scaladoc above
extension, which dodges the same trap, so it is untouched.

Verified rather than asserted: javap -p -c over JsonOptics$package$ and
SchemaOptics$package$ is byte-for-byte identical before and after the
bracing, and zioIntegration/test passes unchanged (268 examples).

kyo's numbers moved too, because #111's survivor-killing pass landed in
between: 20 K / 4 S / 38.5% became 30 K / 5 S / 47.6%.

Evidence: the generated rows

Produced by the patched generator reading the real report.json of a run on
this tree, not written by hand:

| `zio` | 37 | 0 | 6 | 0 | 0 | 86.0% | 86.0% |  |
| `kyo` | 30 | 0 | 5 | 28 | 0 | 47.6% | 85.7% | The no-coverage block is all `RecordIsoMacro`: quoted-macro code that expands at compile time, so like `generics` its mutants leave no runtime footprint. The covered score is the one that reads the hand-written optics. |

• zio is a clean, high-signal row: 37 killed, 6 survived, and no
  no-coverage block at all. Everything stryker can mutate — the DynamicValue
  and zio.json tree navigation, the JsonCursor write descent, the Chunk
  index guards — is genuinely executed by the suite. It needs no Notes caveat,
  so it has none.
• kyo's low total score is a macro artefact, not a coverage hole. All 28
  no-coverage mutants are in RecordIsoMacro — quoted-macro code that expands at
  compile time and leaves no runtime footprint for a test run to cover, the
  same structural reason generics scores 0%. The covered column (85.7%) is
  the one that describes the hand-written optics, and the Notes column says so.

Only the mutation table's two new rows are touched; the coverage table and the
composition matrix are left byte-identical, because a full in-place generator run
in a clone without a scoverage aggregate would have replaced the real coverage
numbers with "No scoverage aggregate report found".

The n/a branch stays, without zio as its example

Nothing hits it today. It is kept anyway: the old code rendered a zero-mutant
module as the same em-dash it uses for "no report at all", conflating "nothing
to score" with "nothing ran", and printing 0.0% would read as a test-quality
failure for a module that offers a mutator no purchase. n/a is also what
stryker's own console prints. The page now states the distinction as a general
rule instead of pointing at zio.

The all-NoCoverage shape is deliberately left alone: generics keeps 0.0% total
with an em-dash covered score, because those mutants do exist — they are simply
never executed.

report shape	total	covered
no mutants at all	n/a	n/a
only Ignored (StringLiteral-excluded) mutants	n/a	n/a
all NoCoverage (the generics shape)	0.0%	—
ordinary kill/survive	86.0%	86.0%
no report.json	—	— (+ the "run sbt mutationAll" note)

Conflict resolution against main

Both conflicting hunks were keep-both — main and this branch edited the same
regions for unrelated reasons.

• site/tools/gen-qa-report.py, comment block: kept test: 29 mutants killed in −2 net test lines — jsoniter oracle rewrite, circe bounds property, core exists/Index, and the equivalent-mutant catalogue #111's reworded
  "a note that claims a module is unscoreable MUST be retired" paragraph and
  this branch's directory-vs-project-id comment above it.
• site/tools/gen-qa-report.py, MUTATION_MODULES: kept test: 29 mutants killed in −2 net test lines — jsoniter oracle rewrite, circe bounds property, core exists/Index, and the equivalent-mutant catalogue #111's retired
  avro/jsoniter notes (the old "Not scored:" strings are gone for good) and
  appended the zio/kyo entries after them.
• site/docs/quality-assurance.md, mutation table: same — test: 29 mutants killed in −2 net test lines — jsoniter oracle rewrite, circe bounds property, core exists/Index, and the equivalent-mutant catalogue #111's corrected
  avro/jsoniter rows plus the two new ones.
• site/docs/quality-assurance.md, prose: main and this branch had each grown a
  zio/kyo explainer, in different places, saying overlapping things. They are
  merged into one block, keeping fix(kyo): brace the valuePrism extension so stryker4s can mutate the module #109's braced-extension detail from main and
  the covered-vs-total framing from here.
• .github/workflows/quality.yml auto-merged: main's new cost job and this
  branch's stryker-loop edit are disjoint.

Nothing from either side was dropped. #111's Known equivalent mutants
catalogue is untouched.

Gates

Clone on Temurin 25, never the primary repo. kyo confirmed in the aggregate.

• sbt "++ 3" compile test — success, 0 errors; kyo compiled and its four
  suites ran.
• sbt benchmarks/compile — success (outside the root aggregate, breaks
  silently).
• sbt mimaReportBinaryIssues — success.
• sbt "docs/mdoc; docs/laikaSite" — success; rendered quality-assurance.html
  carries both new rows.
• sbt scalafmtCheckAll "scalafixAll --check" — success, run last, after
  purging the target/streams/*/scalafmt* and *scalafix* caches; it genuinely
  re-read 20 source groups rather than reporting a cached pass.
• zioIntegration/test — 268 examples, 0 failures, with the braces applied.
• project zioIntegration; stryker / project kyoIntegration; stryker, in the
  project-switch form — never <module>/stryker, which reports 100% NoCoverage.
• javap -p -c diff over the two affected package objects — identical.

🤖 Generated with Claude Code

https://claude.ai/code/session_0194EHFR4NamCpTHiqy7B74V

[github-actions]

🚀 Cloudflare Pages preview for fix/qa-report-zio-kyo-modules is live:

https://43c69ef7.cats-eo-docs.pages.dev

Branch alias: https://fix-qa-report-zio-kyo-module.cats-eo-docs.pages.dev

_{Built from commit eee9a1272fb81dd638c0c6ccfc0c8eead34a3c1d · updated on every push.}

[github-actions]

Benchmark A/B

Allocation (B/op) — authoritative

Benchmark	params	base	head	Δ
ZioDiBench.eoDrillGet	-	0.0	0.0	-1.2%
ZioDiBench.eoServiceGet	-	0.0	0.0	-0.0%
ZioDiBench.eoDrillModify	-	528.0	528.0	+0.0%
ZioDiBench.eoServiceReplace	-	456.0	456.0	-0.0%
ZioDiBench.handDrillModify	-	648.0	648.0	+0.0%
ZioDiBench.handServiceGet	-	120.0	120.0	-0.0%
ZioDiBench.handServiceReplace	-	576.0	576.0	-0.0%
ZioDiBench.handDrillGet	-	120.0	120.0	+0.0%

Timing (ns/op) — directional only, same-VM but shared runner

Benchmark	params	base	head	Δ
ZioDiBench.eoDrillModify	-	78.3	84.0	+7.2%
ZioDiBench.eoServiceReplace	-	62.6	60.5	-3.3%
ZioDiBench.handDrillModify	-	71.0	72.2	+1.7%
ZioDiBench.eoDrillGet	-	6.8	6.8	-1.1%
ZioDiBench.handServiceGet	-	8.6	8.5	-1.0%
ZioDiBench.handServiceReplace	-	64.4	64.1	-0.5%
ZioDiBench.handDrillGet	-	8.4	8.4	+0.5%
ZioDiBench.eoServiceGet	-	2.9	2.9	-0.4%

_{base_sha: ad30b87ce2171fdc815fc72e8cb6e4eba6c988d7 · head_sha: eee9a1272fb81dd638c0c6ccfc0c8eead34a3c1d · jdk: temurin-21 · runner: ubuntu-22.04 · jmh_params: -i 3 -wi 2 -f 1 -t 1 -foe true -prof gc -rf json · profile: pr:-i3-wi2-f1-t1-gc}