-broke out the provider-specific llm_funcs into 4 separate files:
1. stream,
2. response
3. conversation
4. embeddings
-deprecated NPCSH_MODEL and NPCSH_PROVIDER in favor of NPCSH_CHAT_MODEL and NPCSH_CHAT_PROVIDER and more specifically integrated in other ENV variables for reasoning, image gen, image analysis, and embeddings.
-fixed a couple bugs here and there (including the one identified in 80 )