feat: Add large codebase support for MCP server

[tnm]

Summary

Addresses GitHub Issue #37 - MCP server timing out on large codebases (Linux kernel, Kubernetes, etc.).

Changes:

• Increase grep timeout from 30s → 120s (configurable via KIT_GREP_TIMEOUT env var)
• Add grep early termination using -m flag for faster results
• Add get_file_tree pagination with limit/offset params (default 10K files per page)
• Add warm_cache MCP tool to pre-warm caches before heavy operations
• Add 16 tests for new features
• Update MCP docs with large codebase guidance

Benchmarks (Linux kernel - 98K files, 1.9GB)

Operation	Before	After	Improvement
File tree (cold)	N/A	3.65s	baseline
File tree (cached)	N/A	0.015s	248x faster
Grep (max 100 results)	timeout at 30s	18s	✅ completes
Pagination	not supported	✅ works	unlimited repo size

Key Design Decisions

1. warm_cache is opt-in: Users call it explicitly before heavy operations (not automatic)
2. Pagination defaults to 10K files: Enough for most repos, prevents context explosion
3. Grep timeout configurable: Set KIT_GREP_TIMEOUT=300 for very large repos
4. All speedups work without env vars: Settings are optional overrides

Test plan

• All 16 new tests pass
• Existing test suite passes (197 tests)
• Verified on Linux kernel (98K files)
• Verified on Kubernetes (28K files)