feat: add type-aware class-member proof of concept

[BartWaardenburg]

Summary

• add an experimental dead-code --type-aware refinement for unused class members
• keep extraction, graph construction, caching, and default output syntactic-only
• use a backend-neutral protocol with a reference TypeScript-Go 7.0.2 sidecar
• emit bounded provenance, warnings, timings, and explain metadata under _meta.type_aware

The design follows the architecture validated by Oxc's stable type-aware linting release: the Rust analyzer keeps its existing responsibilities while a separate semantic engine builds TypeScript programs and answers narrow type queries.

Result

• the nested generic reproduction reports both used and dead syntactically, then only dead with --type-aware
• on Astro 6.3.1, the sidecar examined 104 candidates, confirmed 38 exact uses, conservatively retained 66, and added no findings
• the final Astro smoke retained the genuinely unused NodePool.clear candidate
• the semantic phase took approximately 1.1 to 2.7 seconds across measured local runs; no overall speedup is claimed

Safety

• only exact declaration matches can remove a finding
• diagnostics and inferred projects remain conservative
• malformed, incomplete, oversized, or incompatible protocol responses fail with exit 2
• blocked I/O, timeouts, output limits, Ctrl+C, and SIGTERM terminate the full sidecar process tree
• the normal command never discovers or starts the sidecar

Verification

• nested generic reproduction
• TypeScript-Go sidecar behavior and inferred projects
• Rust protocol, lifecycle, CLI, and output-contract coverage
• Astro and additional real-project smoke tests
• deterministic repeated output
• cold, warm, and peak-memory measurements
• generated schema and TypeScript contract validation
• workspace check, tests, strict lint, format, rustdoc, typo, workflow, and self-audit gates
• native Rust, CLI-output, and JSON-contract review

Scope

This is a proof of concept. Production packaging, persistent semantic caching, MCP and editor integration, and additional type-aware issue families remain separate product decisions.

[codspeed-hq]

Merging this PR will not alter performance

✅ 49 untouched benchmarks
⏩ 21 skipped benchmarks¹

---

_{Comparing feat/type-aware-class-members-poc (79e94d1) with main (45f1642)}

Footnotes

1. 21 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩