Sample Curator v0.1.6

Sample Curator v0.1.6 — LLM naming refinement backend selector

The LLM-refined filename feature now offers three explicit backends.
Pick one in Settings → Analysis & ML → LLM naming refinement:

- **Apple Foundation Models** — zero install, runs on-device.
  Requires macOS 26.0+, Apple Silicon, Apple Intelligence enabled.
- **Ollama** — local daemon, fast quantized inference.
  Requires `brew install ollama` + `ollama pull <model>`.
- **HuggingFace Transformers** — runs in the bundled sidecar.
  First use downloads ~1 GB for Qwen 2.5 0.5B Instruct.

Existing installs with `model_id: "ollama:..."` migrate silently.

Also: bundled .app launches cleanly on macOS 26 (the previous build
referenced @rpath/libswift_Concurrency.dylib which couldn't resolve).

Closes vault-3ume + vault-15i9.