Mortar ships local FIM autocomplete

Mortar is small by design, which is exactly the point: local-code-completion users often need less product surface, not another full AI IDE.

• –The setup flow is built around endpoint and model selection, making it a cleaner companion for `llama-server` and `llama-swap`.
• –Native `/infill` support matters because FIM-capable coding models can use prefix and suffix context without prompt-template guesswork.
• –The OpenAI-compatible fallback broadens backend support while preserving the same autocomplete-only UX.
• –This is early and tiny, with a v0.1.0 release and minimal community signal, but it targets a real pain point in local AI coding workflows.