Update expert mode to use diffusiongemma model and include read_b64 helper

Update expert mode LLM to mistral-medium-3.5-128b with reasoning effort

Update expert mode routing for sarvam-m

Manual update to nvidia_llm

UI tweak for feedback area and fix token usage metrics extraction

Implement Dual LLM Toggle: Expert and Assistant modes

Update nvidia_llm to use google/diffusiongemma-26b-a4b-it via requests

Update nvidia_llm to use minimaxai/minimax-m2.7 and new sampling params

Update nvidia_llm model, endpoint, stream=False, and remove hardcoded API key

Fix nvidia_llm httpx connection leaks and improve crew.py retrieval logging

Replace gen_llm with nvidia_llm and add frontend test