Add local semantic meeting search with a pinned embedding runtime

[rteoo]

Summary

• Add opt-in local semantic search for meeting transcripts, with model installation, indexing, cancellation, and keyword-search fallback.
• Build and verify the approved source-pinned llama.cpp runtime in CI and release packaging.
• Document runtime setup and update translations, cache relocation, and release probes.

Testing

• CI builds the pinned runtime and runs the embedding runtime probe.
• Windows and macOS bundle workflows run staged embedding probes before promotion.
• Not run (not requested)