feat: add PreToolUse search enrichment via code-review-graph enrich

[gzenz]

Summary

When agents use Grep/Glob/Bash(rg/grep)/Read, a Claude Code PreToolUse hook automatically enriches results with structural context from the graph. Zero-friction adoption -- agents get structural context passively without calling graph tools explicitly.

• New enrich.py module with pattern extraction and graph queries
• New enrich CLI subcommand (reads hook JSON from stdin)
• Updated hooks config in skills.py (Grep|Glob|Bash|Read matcher)
• Platform-split instructions: CLAUDE.md gets lighter instructions focused on deep-analysis tools (hooks handle exploration); non-hook platforms (.cursorrules, AGENTS.md, GEMINI.md, .windsurfrules) keep stronger "prefer graph tools" guidance
• 24 new tests in test_enrich.py
• Updated docs: CHANGELOG, CLAUDE.md, COMMANDS, FEATURES, LLM-OPTIMIZED-REFERENCE, ROADMAP, USAGE

What the hook injects

For each search pattern, the enrichment provides:

• Matching symbols with file:line location
• Community membership (which module/subsystem)
• Callers and callees (max 5 each)
• Execution flows the symbol participates in (max 3)
• Tests that cover it (max 3)

For file reads, it provides the same structural context for all functions/classes in the file.

Upgrade note

If upgrading from v2.1.0, delete the <!-- code-review-graph MCP tools --> section from your CLAUDE.md and re-run code-review-graph install to get the updated lighter instructions.

Test plan

• 24 new tests (pattern extraction, search enrichment, file enrichment, hook output)
• Full suite: 589 tests pass
• Lint and type-check clean

[tirth8205]

Major feature — passive graph enrichment via PreToolUse hooks is a compelling idea. New enrich.py module + CLI command + 237 lines of tests.

This needs rebase on latest main (has conflicts). We'll review the enrichment pipeline and hook integration in detail once it's rebased.

[gzenz]

@tirth8205 thanks for your feedback. actually, instead of opening further pr's, i worked locally quite a bit and did many more improvements. for the sake of simplicity, i'll open a new pr based on current master with all of them.