feat(devx): ci.yml and the completeness join understand a file-level shard item

A shard item is now a package or a package plus a k/n slice. turbo applies a
passthrough run-wide, so a slice cannot share an invocation with packages that
own fewer test files than n -- the whole packages keep one turbo run and each
slice gets its own, filtered to the one package it slices. Each leg tees to its
own log because run-with-stall-guard truncates, and the legs are concatenated
even when one failed, which is when the completeness guard earns its keep.

The scheduled join folds items to package names (an unfolded one reaches
describe() as a package the turbo ls document never listed, and the guard
refuses the whole shard over it), and parseRunCompleted folds every roster line
rather than the last, so a completed second invocation cannot vouch for a first
that stopped early.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Vbw3RPgdtqesx4azk9SbW8

Author: claude

.github/workflows/ci.yml

@@ -539,7 +539,7 @@ jobs:
           node scripts/partition-test-shards.mjs "$RUNNER_TEMP/turbo-ls.json" \
             --shard ${{ matrix.shard }}/6 --exclude @objectstack/dogfood \
             > "$RUNNER_TEMP/shard-packages.txt"
-          echo "Packages on this shard:"
+          echo 'Items on this shard (a package name, or a package plus a k/n file-level slice):'
           cat "$RUNNER_TEMP/shard-packages.txt"
 
       # --concurrency=4: turbo's default (10) oversubscribes the 4-vCPU
@@ -587,11 +587,61 @@ jobs:
           # vitest's own "use the default" signal. Needs turbo.json's
           # globalPassThroughEnv entry or turbo strips it — see the script header.
           export VITEST_MAX_WORKERS="$(node scripts/vitest-worker-cap.mjs)"
-          FILTERS=$(sed 's/^/--filter=/' "$RUNNER_TEMP/shard-packages.txt" | tr '\n' ' ')
           mkdir -p "$RUNNER_TEMP/stall-reports"
-          node scripts/run-with-stall-guard.mjs --log "$RUNNER_TEMP/test-core.log" --stall-minutes 10 \
-            --report-dir "$RUNNER_TEMP/stall-reports" -- \
-            pnpm turbo run test $FILTERS --concurrency=4 --summarize --log-order=stream
+
+          # Split the shard's ITEMS into the whole packages, which share one
+          # turbo run as they always have, and the file-level slices, which
+          # cannot: `--shard=k/n` is passed through to vitest by turbo as a
+          # RUN-level argument, so it would reach every package in the run —
+          # and on any package with fewer test files than n that is a hard
+          # vitest failure (or, with --passWithNoTests, silently no tests at
+          # all). A slice therefore gets its own invocation, filtered to the one
+          # package the partitioner sliced.
+          FILTERS=""
+          SLICES=""
+          while read -r PKG SLICE; do
+            [ -n "$PKG" ] || continue
+            if [ -n "$SLICE" ]; then
+              SLICES="$SLICES $PKG=$SLICE"
+            else
+              FILTERS="$FILTERS --filter=$PKG"
+            fi
+          done < "$RUNNER_TEMP/shard-packages.txt"
+
+          # Each leg tees to its OWN log — run-with-stall-guard opens the log
+          # with 'w', so a second leg pointed at one path would truncate the
+          # first leg's output and the completeness guard below would grade half
+          # a shard. They are concatenated afterwards, and that concatenation
+          # happens even when a leg failed, because a red suite is exactly when
+          # the completeness guard earns its keep.
+          #
+          # ⛔ A failing leg STOPS the remaining ones, the same way turbo stops
+          # scheduling on the first failure inside one run. Carrying on would add
+          # a second full suite to a job that is already red and already inside a
+          # 30-minute wall — turning an informative red into a killed job with no
+          # attestation at all, which is the #16173 failure mode itself.
+          STATUS=0
+          LOGS=""
+          for LEG in __whole__ $SLICES; do
+            if [ "$LEG" = __whole__ ]; then
+              [ -n "$FILTERS" ] || continue
+              LOG="$RUNNER_TEMP/test-core-packages.log"
+              set -- pnpm turbo run test $FILTERS --concurrency=4 --summarize --log-order=stream
+            else
+              PKG="${LEG%%=*}"
+              SLICE="${LEG#*=}"
+              LOG="$RUNNER_TEMP/test-core-slice-$(printf '%s' "$PKG" | tr -c 'A-Za-z0-9' '-').log"
+              set -- pnpm turbo run test "--filter=$PKG" --concurrency=4 --summarize --log-order=stream -- "--shard=$SLICE"
+            fi
+            LOGS="$LOGS $LOG"
+            node scripts/run-with-stall-guard.mjs --log "$LOG" --stall-minutes 10 \
+              --report-dir "$RUNNER_TEMP/stall-reports" -- "$@" || { STATUS=$?; break; }
+          done
+
+          if [ -n "$LOGS" ]; then
+            cat $LOGS > "$RUNNER_TEMP/test-core.log"
+          fi
+          exit $STATUS
 
       # --summarize above costs nothing at runtime and writes
       # `.turbo/runs/<id>.json`: one per-task record with the execution window