Cleanup: remove internal handoff docs, BUILD.md, and dev eval logs (keep schemas/ runtime dep)

GPU tuning: @spaces.GPU duration 90->60, max_new_tokens 768->320, remove diagnostic logging

DIAGNOSTIC: log the swallowed transformers-backend exception (full traceback)

Fix dispatcher fumbles: raw_decode first JSON object (small model repeats/think-wraps it; greedy regex spanned multiple braces -> no-op)

Deploy Mumbai Local Panic — Nemotron(ZeroGPU) dispatcher + VoxCPM2 voice