perf: resolve scan/review pipeline performance bottlenecks (P2)

[ajianaz]

Summary

Fixes 10 performance bottlenecks across the scan/review pipeline. All are "repeated work in loops" patterns — the fix is always the same: do it once outside the loop.

Fixes

#	File	Issue	Fix
#41	src/engine/rules/matching.rs	Regex recompiled per line	Pre-compile regex once via ensure_compiled()
#5	src/mcp/tools.rs	Per-file DB queries in nested loops	Batch all symbols in single IN query
#13	src/main.rs	Repeated graph lookups in Affected cmd	Single batch query for all changed files
#14	src/main.rs	Prepared statement in inner loop	Prepare once before loop
#59	src/engine/secrets_scanner.rs	No early cutoff after max_findings	Break out of all loops when cap reached
#70	src/engine/context/extraction.rs	Symbol cap only breaks inner loop	Break all three nested loops
#67	src/engine/context/resolver.rs	Repeated file reads from disk	HashMap<PathBuf, String> cache
#3	src/mcp/tools.rs	New Tokio runtime per MCP call	Static LazyLock<Runtime>
#22	src/index/extract.rs	Regex per detect_function_entry call	static LazyLock<Regex> per language
#24	src/index/mod.rs	Per-file transaction in prune_deleted	Single transaction for all deletes

Testing

• ✅ All 609 tests pass (cargo test)
• ✅ Clippy clean (cargo clippy)
• ✅ Formatted (cargo fmt)

Refs #335