Add managed local llama.cpp embeddings

[mika76]

Summary

• add a managed local llama.cpp embedding provider with globally managed runtime and model assets
• add curated managed model install/list/use/remove flows plus model-id shell completions
• gate the managed runtime by supported platform and reuse the runtime via health checks instead of PID probing

Notes

• the managed llama.cpp runtime is pinned to upstream build b3426
• manual end-to-end testing was only done on macOS
• on unsupported platforms, the managed llama.cpp functionality is hidden or returns a clear unsupported-platform error rather than pretending to work

Validation

• go test ./config ./embedder ./internal/managedassets ./search ./indexer ./cli
• manual smoke test on macOS: init, model install, model use, watch, search