[Python] tornado client does not support chunked responses

Description

The python tornado client does not support chunked/streaming responses. When _preload_content is True the entire response body is read. When _preload_content is False the response body is still read but the raw tornado request is returned instead (probably a bug).

I think the expected behavior should be similar to using urllib3: When _preload_content=True the response body should not be be read, allowing the caller to call read_chunk on the response object, and presumably handle the deserialization of the chunks themselves.

Suggest a fix/enhancement

Unfortunately, for tornado to support reading streaming responses you must provide a streaming_callback to the request which is called whenever a chunk arrives off the socket. There is no other way way to prevent the client from attempting to read the entire response body. Allowing the caller to specify a callback doesn't really mesh well with the way the APIClient works, so it likely requires some trickery to make it work (and a bit of buffering).

FWIW, the python asyncio client probably has a similar issue where the response body is always read, though the fix is probably simpler since aiohttp doesn't require a callback, but I've not looked into this much.

I have a PR for this in progress.

Start with the Python Tornado client and APIClient handling of _preload_content, comparing the behavior described with urllib3 and the asyncio client. Trace how Tornado requests consume response bodies and how a streaming callback could be buffered. Done means chunked responses can be consumed incrementally without forcing the full body read, while preserving the expected deserialization behavior.