safely interrupt streaming requests

[thehunmonkgroup]

Enables raising an error from a callback handler to interrupt streaming requests.

Handles cleanup of the underlying HTTP request.

Here's an example usage of a callback handler that I've tested as working:

class VerboseStreamingStdOutCallbackHandler(StreamingStdOutCallbackHandler):
    @property
    def always_verbose(self) -> bool:
        """Whether to call verbose callbacks even if verbose is False."""
        return True

def make_interrupt_streaming_callback_handler(backend):
    class InterruptStreamingCallbackHandler(VerboseStreamingStdOutCallbackHandler):
        def on_llm_new_token(self, token: str, **kwargs: Any) -> None:
            if not backend.streaming:
                message = "Request to interrupt streaming"
                backend.log.info(message)
                raise EOFError(message)
    return InterruptStreamingCallbackHandler()

I have not investigated the _agenerate method, or any other LLMs yet -- would like to get some feedback/confirmation on the approach first.