Harden, fix, and enhance Code Historian across capture, search, and tooling

[KirtiJha]

Correctness fixes:

• capture: saves no longer silently drop changes (fall back to last-known
  content); bound the in-memory document snapshot cache
• metadata: wrap batch inserts in a transaction; gzip-compress diffs at rest
  and decompress transparently on read; detect/rebuild corrupt databases on
  load; cache stats (60s) to avoid full-table scans
• vectorStore: escape quotes in delete predicate and cap vector topK
• restoration: implement rename restoration
• timeline: fallback diff now shows deletions too (proper LCS diff)
• chat: fix broken module re-export

Enhancements:

• Zero-config built-in embedding provider (now default; deterministic hashing
  embedding, no API key/model/server needed)
• Real BM25 keyword ranking (candidate retrieval + BM25) instead of unranked LIKE
• Local BM25 reranker (free, offline) enabled by default
• Symbol time-travel command + editor context menu entry
• Bookmarks (pin changes) backed by a new table
• Automatic retention enforcement on startup (prune changes + vectors, vacuum)
• Encrypted SecretStorage for API keys with migration out of settings.json
• Prompt-injection hardening for retrieved RAG context
• Markdown rendering: links + fenced code blocks; safer emphasis parsing
• Keybindings (Open Timeline, Search History)

Tooling & docs:

• Vitest unit tests (38) for pure modules + GitHub Actions CI
• Fix broken eslint config; resolve all lint errors
• Create missing walkthrough media; add CONTRIBUTING.md and CHANGELOG.md
• README accuracy: BM25 (not FTS5), correct keybindings/config, qualified perf
• Remove leftover console.log debugging from the webview