Release v0.2.0: task context, indexing resilience, and public docs

[pranitmodi]

Summary

• Adds get_task_context and inspectable hybrid retrieval with exact-symbol ranking, relationship expansion, confidence-based fallback, and a labeled quality benchmark.
• Makes automatic indexing more resilient with per-file watcher updates, content-sample freshness checks, persisted watcher errors, TypeScript aliases, and Python import expansion.
• Prepares the open-source v0.2.0 release with public architecture, benchmark, security, conduct, changelog, and contributor documentation.
• Updates LanceDB/Transformers to clear the production dependency audit and adds code-intel --version.

Measured result

Fresh eight-task run on the final pre-release index:

• 73.9% average per-task token reduction vs workspace scan
• 0.85 relevant-file coverage@5
• 0.96 Recall@10
• 0.84 MRR
• 0.72 NDCG
• 708 ms p95 across measured operations

Verification

• npm run typecheck
• npm test — 132 passed, 11 environment-dependent skipped
• npm run build
• npm audit --omit=dev — 0 vulnerabilities
• npm run benchmark:retrieval
• npm pack --dry-run --json
• Install packed tarball and smoke-test code-intel --version / --help

okay.md is a local note and is intentionally excluded.