FEATURE: Add support for Mistral models

[SamSaffron]

Adds support for mistral models (pixtral and mistral large now have presets)

Also corrects token accounting in AWS bedrock models

[tyb-talks]

Should the provider_id be overridden? As-is, this currently returns the one for OpenAi.

[SamSaffron]

oh sure ... good catch will fix

[tyb-talks]

Minor question on provider_id for the endpoint, but otherwise looks ok.

[tyb-talks]

I just spotted something I missed earlier in the OpenAi endpoint:

        def prepare_payload(prompt, model_params, dialect)
          payload = default_options.merge(model_params).merge(messages: prompt)

          if @streaming_mode
            payload[:stream] = true

            # Usage is not available in Azure yet.
            # We'll fallback to guess this using the tokenizer.
            payload[:stream_options] = { include_usage: true } if llm_model.provider == "open_ai"
          end

[tyb-talks]

😅 actually it's fine, it just omits those params if it's not from OpenAi. 🤔 If the logic in the inherited endpoint changes more specifically for OpenAi though, it would also affect this Mistral endpoint. Still fine for now.

[SamSaffron]

Yeah this is safe by "side effect" cause mistral just sends this info anyway.

[SamSaffron]

thanks for the review @tyb-talks 🤗