Sync cached_property getter access

[mjpieters]

Add an option to pass in a lock to the cached_property decorator, so the
accessor is run just once even when multiple tasks are concurrently trying to
await on the value.

This PR fixes concurrent access to the property getter. Notable changes include:

• The cached_property decorator now accepts either a getter coroutine method,
  or an async context manager. In the latter case a new decorator is returned
  to accept the actual getter.
• Aligned cached_property internals with the stdlib.functools.cached_property version. Most notably, the test for instance.__dict__ has been moved to __get__, and the __wrapped__ and _name attributes now match the attribtues on a stdlib cached property descriptor.
• Improved the type annotations; e.g. the return type of the awaitable getter is covariant.

Closes #149

[maxfischer2781]

The CodeQL "Statement has no effect" reports for ellipsis all are false positives and can be ignored: github/codeql#11351

[maxfischer2781]

Thanks for this nice work! I need some time to review it, but at a glance I don't see any glaring problems.

[maxfischer2781]

Thanks for the changes. That is some nicely thought out code, I had to go back a few times when realising I misunderstood an edge case that you considered.

Aside from some minor things, there's three main points that I stumbled on:

• I don't think the fast path in _await_impl is worth the code size cost. Please correct me if I'm missing a reason to keep it.
• The locked section in _await_impl could drop the lock more aggressively.
• I'm not happy with either variant of the new parameter name. Unless you have an idea for a better one, I prefer to not expose it by making the parameter positional-only.

[mjpieters]

This should be ready for re-review now.

[maxfischer2781]

The static check failures are unrelated and will be fixed by #152, no need to duplicate them here.