Speed up long-context prefill on Apple10 (drumih#16)

* Speed up long-context prefill on Apple10

* Condense TensorOps benchmark wording

* Document Apple10 prefill assumptions

---------

Co-authored-by: Andrey Mikhaylov <andreymikhaylov@Andreys-MacBook-Pro.local>

Add slim resumable model downloads (drumih#14)

Add OpenAI-compatible local server and tool calling (drumih#13)

* Add OpenAI-compatible local server

* Fix graceful server shutdown

---------

Co-authored-by: Andrey Mikhaylov <andreymikhaylov@Andreys-MacBook-Pro.local>

Make the system design easier to follow (drumih#5)

* Make the system design easier to follow

* Remove audio references from public docs

---------

Co-authored-by: Andrey Mikhaylov <andreymikhaylov@Andreys-MacBook-Pro.local>

App UI fixes (drumih#3)

* Fix Mac runtime controls presentation

* Explain temperature control

* Refresh app screenshot and skip draft CI

---------

Co-authored-by: Andrey Mikhaylov <andreymikhaylov@Andreys-MacBook-Pro.local>