Chat: uplift errors of model on user requests

[tiero]

We can start from Mistral binaries, for example instead of returing a general Load failed we should return the ValueError below

    | Traceback (most recent call last):
    |   File "starlette/responses.py", line 273, in wrap
    |   File "starlette/responses.py", line 262, in stream_response
    |   File "routes.py", line 64, in generate_chunk_based_response
    |   File "llama_cpp/llama.py", line 1540, in _convert_text_completion_chunks_to_chat
    |   File "llama_cpp/llama.py", line 947, in _create_completion
    | ValueError: Requested tokens (3308) exceed context window of 512
    +------------------------------------