fix(honua-gp): bind Result outputs to real artifacts, reject unsupported parameter semantics

[mikemcdougall]

Problem

run_layer_process (the shared entry point for analysis.Buffer,
analysis.SpatialJoin, management.Dissolve, management.Project)
registered a tool's requested output name as a session alias before the
job ran, resolved against the literal arcpy name via
resolve_or_register_output / descriptor_mapping. That resolution had no
relationship to what the server actually created and could default to
layerId=0 of an unrelated dataset. A later GetCount / SearchCursor /
chained GP call against that output name could therefore silently read the
wrong data instead of the tool's real result.

Separately, Buffer/SpatialJoin/Dissolve/Project accepted (and
ignored) several arcpy parameters that change tool semantics
(line_side, join_operation, multi_part, preserve_shape, ...) without
telling the caller.

Fix

• Reserve, then bind. The output name is only reserved before
  submission (a fail-fast duplicate check honoring
  env.overwriteOutput, mirroring arcpy) — no alias is published yet. After
  the job reaches successful, the real output artifact is extracted from
  the job's results document and that identity is what the alias, and
  therefore every later Result[0]/getOutput(0)/str(result)/downstream
  call, resolves through. A missing/empty output on a "successful" job now
  raises a typed ExecuteError(error_kind="missing_output") instead of
  returning a successful Result.
• No accidental rollback needed. Since the alias is never mutated before
  success, a failed/dismissed job leaves any prior alias under that name
  completely untouched — no separate rollback bookkeeping required.
• Reject unsupported parameter semantics. Every arcpy_params entry the
  projector doesn't translate into a process input is now validated:
  nondefault line_side/line_end_type/method (Buffer),
  join_operation/join_type/field_mapping/distance_field_name
  (SpatialJoin), statistics_fields/multi_part/unsplit_lines (Dissolve),
  and transform_method/in_coor_system/preserve_shape/max_deviation/vertical
  (Project) now raise HonuaGpConfigurationError before submission instead
  of being silently dropped.
• Fixed a latent NameError. resolve_layer_id raised
  HonuaArcpyResolveError, a name that doesn't exist in _errors.py
  (HonuaGpResolveError is the real class). This fired for any
  honua://services/<svc> URI with no layer segment — untested until now.

Evidence

• test_output_binding.py (new): drives a full
  Buffer -> GetCount/SearchCursor -> Dissolve chain through a fake
  transport whose feature catalog is keyed by the exact resolved source
  identity, with deliberately different feature counts/geometry at each
  stage (5 line inputs -> 2 buffered polygons -> 1 dissolved polygon). Also
  covers a subsequent read against an expired/gone artifact surfacing as a
  typed ExecuteError.
• test_process_tools.py: new coverage for every rejected/accepted
  parameter combination per tool, a dismissed job, an empty-outputs
  "successful" job, fail-fast duplicate-output rejection (no job
  submitted), and a failed overwrite leaving the prior alias untouched.
• test_resolve.py: regression test for the resolve_layer_id
NameError.
• docs/honua-gp/compatibility-matrix.md /
  packages/honua-gp/docs/compatibility-matrix.md regenerated from the
  updated _compat.py notes (kept byte-equal per the existing drift gate).

Full existing suite plus the new tests: 226 passed, 4 skipped. ruff check .
clean. honua-gp eval harness: 50/50 (100%).

Scope note on AC #4 / #6: this repo's test suite has no live
honua-server available (the opt-in live-server lanes —
tests/conformance, ephemeral-server-smoke, staging-smoke — need
infrastructure this sandbox doesn't have, and both smoke lanes are
currently failing on every open PR in this repo for unrelated,
pre-existing reasons: an expired honua-demo-infra staging binding and
the ephemeral seeded server not reaching /healthz/ready, neither a
required status check). The chained proof above is the strongest
client-SDK-side evidence available here: it fully exercises projection,
alias-binding, and downstream resolution against a controlled transport
using the same fake-OGC pattern every other test in this package already
uses. A genuine real-server value/geometry proof, and the cross-repo
installed-consumer evidence coordination with #202/#205/#196, are left to
land once that infrastructure is reachable from CI.

Refs #226 (released: real-server execution proof and #202/#205/#196
cross-repo evidence coordination need live honua-server infrastructure not
reachable from this repo's CI today; every other acceptance criterion —
output binding correctness, typed missing-output failure, overwrite/alias
preservation, parameter-semantics validation, and the resolve_layer_id
NameError — is fixed and covered above)

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.

[chatgpt-codex-connector]

You have reached your Codex usage limits for security reviews. Please try again later.