Fix potential deadlock in PyFutureAwaitable callback dispatch

[ColemanDunn]

fixes #750

Reproduction of the issue can be found at: https://github.com/ColemanDunn/granian-channels-lock-up

Change is to not hold the RwLock across call_soon_threadsafe. The comment is somewhat self documenting of what this change does. I can take it out if needed. Perhaps there is a better fix than this in the underlying code, but I am now unable to reproduce the lockup.

[gi0baro]

I'm prone to consider this as a random effect, other than the actual solution.

I'm not exactly sure how the RwLock from PyFutureAwaitable.ack can led to deadlocks, given:

• all the calls to code interacting with that lock are performed holding the GIL, thus:
  • none of the involved code can actually access that lock concurrently
  • the main thread is detached from the GIL (with the event loop in idle) when set_result gets called
• call_soon_threadsafe doesn't really block, it simply add the callback to the loop "queue" and write a wake event

I'm going to still review the code and possibly merge it, as it might be a valid concern for the free-threaded builds, but again, it's hard to think this can be the actual solution on whatever the problem in #750 is, and more like a coincidence.