feat: parser refactoring, Jedi call resolution, and performance optimizations

[gzenz]

Summary

Major improvements spanning parser architecture, call graph accuracy, build performance, and dead code detection:

• Parser refactoring: Extracted 16 per-language handler modules into code_review_graph/lang/ using a strategy pattern, replacing monolithic conditionals in parser.py. Thread-safe parser caches with double-check locking.
• Jedi-based call resolution: New jedi_resolver.py resolves Python method calls at build time. Pre-scan filtering by project function names reduces enrichment from 36s to 3s on large repos. New [enrichment] optional dependency group.
• PreToolUse search enrichment: New enrich.py module and code-review-graph enrich CLI command inject graph context (callers, flows, community, tests) into agent search results passively via hook.
• Call graph improvements: Typed variable call enrichment (Python, JS/TS, Kotlin/Java), star import resolution, namespace imports, CommonJS require(), Angular template parsing, JSX handler tracking, module-qualified call resolution, function/class references as arguments.
• Dead code FP reduction: Framework decorators recognized as entry points, CDK construct methods and abstract overrides excluded, e2e test directories filtered.
• Community detection 21x speedup: Bulk node loading + adjacency-indexed cohesion computation (48.6s to 2.3s on 41k-node repos).
• Build performance: Batch file storage (50-file transactions), batch risk_index (2 GROUP BY queries replace per-node loops).
• DB schema v8: Composite edge index for upsert performance (v7 reserved by PR fix: add sqlite edge compound indexes #127).
• Other: Weighted flow risk scoring, transitive TESTED_BY, --quiet/--json CLI flags, search deduplication, 829+ tests (up from 615).

Evaluation

Tested against Gadgetbridge (41k nodes, 280k edges):

• 8/10 scorecard PASS (callers_of, callees_of, tests_for, communities, flows, impact radius, risk scores)
• Call resolution rate improved from 28% to 39.6%
• Community detection: 48.6s to 2.3s
• Full build time: ~22s for 3,574 files

Migration note

Our composite edge index migration is numbered v8 to avoid conflict with v6 (summary tables, already on main) and v7 (reserved by PR #127).

Test plan

• uv run pytest tests/ --tb=short -q -- 829 passed, 4 skipped
• uv run ruff check code_review_graph/ -- all checks passed
• Full rebuild + evaluation on Gadgetbridge (41k nodes)
• Full rebuild + evaluation on internal Python/TS/React project (261 files)
• CI pipeline (lint, type-check, security, test matrix)