Add an offline terminology snapshot cache and versioned loader #884

Summary

Loading large free vocabularies on every run is slow, and repeated terminology-server lookups are unnecessary for stable concepts. A versioned, content-addressed offline snapshot cache lets grounding load indexes fast and reuse prior lookups, fully local-first.

Scope

• Add clinical/grounding/snapshot_cache.py with a versioned, content-addressed on-disk cache for built vocabulary indexes
• Store index artifacts plus a manifest with system URI, release version, and content hash
• Provide load-or-build semantics and integrity verification on load
• Expose a cache directory honoring OPENMED_OFFLINE and a clear miss/hit API

Acceptance criteria

• First load builds and persists a snapshot; second load reads it without rebuilding
• Corrupted or hash-mismatched snapshots are rejected and rebuilt
• Cache respects the offline flag and stores no raw PHI or restricted concept text by default
• Snapshot manifest is reproducible and version-pinned

Out of scope

• Downloading vocabulary releases (loaders handle that)
• Caching any restricted terminology content without explicit user opt-in

Files

• openmed/clinical/grounding/snapshot_cache.py
• tests/unit/clinical/grounding/test_snapshot_cache.py

Task: OM-527 · Milestone: v2.2 · Priority: P1 · Size: M
Depends on: OM-510 · Blocks: —
Roadmap: v2.2 trustworthy clinical data exchange follow-on wave
Spec: PLANS/V2/EXECUTION/tasks/OM-527.md