Barry Cache v0.4.0

Barry Cache now reuses parsed context snapshots between retrieval commands, reducing repeated reads of the same context files during agent sessions.

Highlights

Added persistent context snapshot caching in .context-cache/context-index.json.
Reused cached context for route, search, load, resume, and review --json.
Added source manifest validation so cached context is refreshed automatically when context files change.
Updated docs and generated agent instructions to describe the new cache behavior.
Bumped package version from 0.3.2 to 0.4.0.

New context cache module:
- Builds a parsed snapshot of feature packs and ADR records.
- Stores the snapshot in .context-cache/context-index.json.
- Tracks source context files and directories using mtimeMs and size.
- Rehydrates cached feature paths relative to the active repo.
- Treats the cache as disposable, so commands still work if the cache cannot be read or written.
New cache regression tests:
- Verifies that Barry creates and reuses a parsed context snapshot.
- Verifies that source file changes invalidate stale cached context.

routeTask, searchContext, loadContext, resumeProject, and review model generation now read through the shared context snapshot loader.
readFeaturePacks now uses the same cached snapshot path for consistency.
README now documents that repeated retrieval commands reuse .context-cache/context-index.json.
Generated Barry agent templates now mention the disposable parsed context index.

Reduced duplicate disk reads during common agent workflows such as:

barry-cache resume --task "..."
barry-cache load --route "..."
barry-cache search --query "..."