Fixes #1909

We'll need to release a final greenlet 2.0 and remove the rc qualifiers here before releasing gevent, because pre-release qualifiers are contagious.

Stop using DEF statements in Cython, as they are deprecated.

See https://github.com/cython/cython/issues/4310

Fixes #1914

Use the non-legacy `setuptools.build_meta` backend.  Prepend the setup
directory to `sys.path` prior to importing modules relative to it.

Fixes #1910

Remove redundant wheel dep from pyproject.toml

Libuv 1.44.2

Run leakchecks on Python 3.11; make them more comprehensive

They're important objects. Locally, the 3.11 tests pass fine this way.

The 2.7 tests had two failures, one from being unable to make a connection in test__server, and one from a timeout in test__doctest. I suspect those are unrelated; lets see what CI says.

Remove the redundant `wheel` dependency, as it is added by the backend
automatically.  Listing it explicitly in the documentation was
a historical mistake and has been fixed since, see:
https://github.com/pypa/setuptools/commit/f7d30a9529378cf69054b5176249e5457aaf640a

Python 3.11rc2 support

By simply not typing it at the Cython level at all. No need since we're directing through functions now.

Fixes #1863
Contributed-By: mgorny

subprocess needed a new check added to check_output, and needed process_group implemented.

because all the leakcheck tests finished successfully after the last commit, and because it sure looks like a refcounting issues in the tree formatting tests (leaked greenlets causing the numbering to be off), try enabling them for this run.

fixes #1872
refs #1867

This should fix the refleaks in general (it was a frame object that was leaking).
Submitting to CI for broad test, and if it is good then we will try backing out the test__util changes too.

Also make it possible for gevent.config to show help.

See https://github.com/gevent/gevent/actions/runs/3205900356/jobs/5239610262

Exiting when idle is because 3.11's test_socket.BasicTCPTest.testDetach() was always failing complaining about a leaked thread. Turns out this thread was from the default threadpool used for DNS resolution.
Something must have changed to cause the first DNS resolution to occur during that test, and hence the first thread to be spawned.  This is opt-in for all thread pools, with the default thread pool opting
in by default; but a config setting allows that to be changed.

Most 3.11 tests pass, except for some in test_subprocess where new functionality was added.

Haven't tried any of the 3.11-specific tests yet.

All tests pass locally on 3.10 except one, reported earlier: test_no_refcycle_through_target (gevent.tests.test__threading_2.ThreadTests)
And a new one that is similar:    test_no_refcycle_through_target (__main__.ThreadTests)

I haven't tried to debug that yet.

Leaving some notes about needed cleanup.

This first push is just to make sure all the other versions besides 3.11 continue to do their thing.

Co-Authored-By: Petr Viktorin <encukou@gmail.com>

test__issue1864.py: Disable on Python 2.7; it failed there for reasons that aren't clear. I verified that Ofast or fast-math was not given. Perhaps some underlying lib had that set?

I'm not very concerned about it though.

This one is available in older versions of Python