fix: restore GET / root route (404 on HF Spaces health check)

fix: lifespan handler, dtype= for transformers 5.x, remove UI, update README

fix: lower token limits for CPU, streaming UI with multi-turn history, /chat instead of /ask

fix: replace torch.cpu.is_bf16_supported() with torch.backends.cpu.get_cpu_capability()

fix: startup event loading, cpu-safe dtype, low_cpu_mem_usage, 503 guard, traceback on error

fix: use dtype= instead of torch_dtype= for transformers 5.x

fix: remove build-time model download — HF secrets unavailable at build time

Remove embedded submodule from index

Initial commit: LFM2-700M FastAPI inference app