[florimondmanca]

Fixes #508

This is an attempt at handling the situation in which users try to use httpx.get() or httpx.Client() in an async environment that has a running loop (e.g. in a Jupyter notebook).

I documented several potential caveats in the form of code comments.

I still need to verify this actually solves the Jupyter use case, though the tests give me hope that it does. :-)

[florimondmanca]

@encode/httpx-maintainers OK, I think I've got something that passes the tests. Basically we're checking if the event loop is running and when relevant run the coroutine in a worker thread instead. This thing is inspired by asgiref's AsyncToSync utility. Definitely needs some cleanup, but at least it seems to work. Also todo: handle shutdown of the threadpool executor (via a new backend.close() method, probably).

[florimondmanca]

Closing as I don't think this will make it, probably in favor of a SyncBackend (unless #522 is accepted) - see #508 (comment).