Consider adding streaming support

[viictorjimenezzz]

It would be very helpful if text responses could be streamed rather than requiring a full await. This feature would be especially beneficial for chat applications. For instance:

• In the case of OpenAI, passing stream=True in the configuration is sufficient.
• For AWS, the converse_stream() function needs to be called instead.

While I'm not a professional developer, I’d like to propose a potential implementation, which builds on the previous issue I raised (#96 ):

#aisuite/providers/openai_provider.py
def response_stream_generator(self, response):
        with response as stream:
            for text in stream:
                yield text.choices[0].delta.content

def chat_completions_create(self, model, messages, **kwargs):
        response = # same

        if kwargs.get("stream", False):
            return self.response_stream_generator(response)
        return response

#aisuite/providers/aws_provider.py
def response_stream_generator(self, response):
        stream = response.get('stream')
        if stream:
            for event in stream:
                chunk = event.get('contentBlockDelta', None)
                if not chunk:
                    continue

                yield chunk.get('delta', {}).get('text', '')

def chat_completions_create(self, model, messages, **kwargs):
        # ... filled inference_config and additional_model_request_fields
        
        additional_model_request_fields.pop("stream", None)

        # Call the Bedrock ConverseStrean API.
        if kwargs.get("stream", False):
            response = self.client.converse_stream(
                modelId=model,  # baseModelId or provisionedModelArn
                messages=formatted_messages,
                system=system_message,
                inferenceConfig=inference_config,
                additionalModelRequestFields=additional_model_request_fields,
            )
            return self.response_stream_generator(response)
        
        response = self.client.converse(
            modelId=model,  # baseModelId or provisionedModelArn
            messages=formatted_messages,
            system=system_message,
            inferenceConfig=inference_config,
            additionalModelRequestFields=additional_model_request_fields,
        )
        return self.normalize_response(response)

See: https://github.com/viictorjimenezzz/aisuite

Thank you for considering this request!

[HOTSONHONET]

@viictorjimenezzz I would like to work on this concept.