[Pipeline: tests-evals] Expand fixtures and regression evaluation for GAMESS

[newtontech]

tracking_id: lsp-perfect-20260615-gamess-lsp-tests-evals

Problem

Top-tier LSP behavior requires broad fixture coverage for valid syntax, invalid syntax, version drift, runtime outputs, and automated fixes.

Target Repository

• Repository: newtontech/gamess-lsp
• Software/backend: GAMESS
• Pipeline stage: tests-evals

Scope

• Add valid and invalid fixtures grounded in official examples and real failure modes.
• Cover diagnostics, hover, completion, symbols, formatting, fix previews, and output/log parsing where applicable.
• Separate parser/unit fixtures from OpenQC/Bohrium smoke fixtures so regressions are easy to localize.

Acceptance Criteria

• Fixtures include positive and negative examples with expected diagnostic IDs and severities.
• At least one regression fixture covers version-specific or runtime/output behavior.
• CI or local scripts run the fixture suite without hidden install steps.

Deterministic Verification

• Run scripts/test.sh or the documented equivalent.
• Run scripts/lint.sh and scripts/typecheck.sh when present, documenting unrelated pre-existing failures.

Implementation Notes

• Follow the repeatable pipeline: official documentation capture -> wiki schema/rules -> provenance -> tests/evals -> LSP runtime/OpenQC integration.
• Keep version-specific behavior explicit. Do not collapse incompatible software versions into one rule.
• Do not claim parity with Python/C++ LSPs until diagnostics, completion/hover/symbols/format/fix surfaces have evidence and tests.
• Link code docstrings to concrete wiki/*.md pages, and link every wiki source claim to a tracked raw/... asset when this stage touches code or docs.