ci: release workflow test gate is dead — continue-on-error forces conclusion=success; flip to outcome + make CI suite honestly green

[rahlk]

PROBLEM

The release workflow's test gate is dead code and has never fired. .github/workflows/release.yml runs tests with continue-on-error: true (line 44) and gates tag deletion on steps.test.conclusion == 'failure' (line 48) — but under continue-on-error a failed step's conclusion is forced to success; only outcome records the failure. Verified on real runs: v1.4.0 and v1.4.1 both had failing tests in Run Tests, conclusion=success, Delete-tag skipped, release published.

SCOPE BOUNDARY

The release workflow's gate, plus the suite hygiene needed to make flipping it safe. Not PR-time CI, which is its own concern. Do not just flip the field — see the caveat.

GOALS

• steps.test.conclusion → steps.test.outcome, so the gate is real. (Landed on release/2.0 via Fix dead release test gate: use outcome, not conclusion #306.)
• Partition the CI suite so "green" is meaningful: env-dependent tests skip cleanly via pytest markers or fixture-presence gating, rather than failing.
• Companion failures resolved — TS golden-fixture collection errors, Java source-code-path crash.
• Same fix applied to main, not only release/2.0.

CAVEATS AND KNOWN RISKS

• Flipping the field alone blocks every release. The gate becomes real against a suite that is not currently CI-green, so the first tag after the flip fails and stays failing. That is why the flip and the suite work are one issue and not two.
• continue-on-error must stay. Removing it fails the job immediately and the tag-deletion step never runs — the opposite of the intent. The two settings are a pair, and Fix dead release test gate: use outcome, not conclusion #306 comments them as such.
• A gate that fires deletes the pushed tag. Worth exercising on a scratch tag before trusting it on a real release.

DEFINITION OF DONE

• Gate fires on a deliberately broken test, verified on a scratch tag or a workflow_dispatch dry run — demonstrated, not reasoned about.
• Release workflow green on an honest, fully-run suite; tests skipped by gating are visibly reported rather than silently absent.
• Both release lines carry the fix.

[rahlk]

Scope consolidation (2026-07-14): this issue now owns the full path to a meaningful CI-green + live gate:

1. Java suite order-dependence — absorbed from test(java): suite is order-dependent — 21/21 in isolation, 15 failures in full-suite runs (global JAVA_HOME mutation / shared cache suspects) #258 (21/21 isolated vs 15 full-suite failures in test_java_sitter; suspects: global os.environ["JAVA_HOME"] mutation in _get_codeanalyzer_exec, shared .codeanalyzer cache, cross-family pollution).
2. TS golden-fixture presence gating — absorbed from test(typescript): golden-fixture tests error instead of skipping when the gitignored fixture is absent (CI + fresh clones) #257 (~23 collection errors in CI/fresh clones must become clean SKIPs or a generated fixture).
3. Source-analysis witness tests — fix(java): source-code analysis crashes with Path(None) — JDK cache resolution assumes a project dir #256 closed as won't-fix (mode unsupported): remove or skip-gate test_get_symbol_table_source_code and test_get_fields_variable_initializers.
4. Only then flip the gate: steps.test.conclusion -> steps.test.outcome, verified against a deliberately broken test on a scratch tag.