Fix Space prefill slowness: cap prefill threads, warm up at startup

Add llama.cpp parser backend for Space

Improve parser runtime config and check-in state

Build training coach MVP