Keep extraction chunks free of context prefixes

Harden provider response handling

Use Nemotron 3 Nano Omni reasoning model

Use NVIDIA hosted Mistral fallback

Support NVIDIA and Mistral provider setup

Fallback to Mistral when Groq is unavailable

Fix query stream startup deadlock

Mark preloaded corpus ready on startup

Add Pluto memory layer and pipeline fixes

Update codebase with latest fixes and improvements

Add application file