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

The release script has been updated to not keep verbose
comments about a past regression. Link to the PR that
describes the regression instead

-pipe tells gcc to use pipes instead of temporary files, which
in general leads to faster compile time, making it a kind of an
optimization flag

I do not reproduce locally on mac, with matching pylint and astroid versions.

See https://github.com/gevent/gevent/actions/runs/3042157082/jobs/5192451149

Fix various build and test failures

```
$ pypy
Python 2.7.18 (596a3d4905d6452b2ac871c1b39dea68418b4f64, May 29 2022, 00:39:20)
...
>>>> from ssl import SSLSyscallError
>>>> SSLSyscallError.__mro__
(<class '_cffi_ssl._stdssl.error.SSLSyscallError'>, <class '_cffi_ssl._stdssl.error.SSLError'>, <class '_socket.error'>, <type 'exceptions.IOError'>, <type 'exceptions.EnvironmentError'>, <typ
e 'exceptions.StandardError'>, <type 'exceptions.Exception'>, <type 'exceptions.BaseException'>, <type 'object'>)
```

Where I think _socket.error is defined:
https://github.com/mozillazg/pypy/blob/release-pypy2.7-v7.3.9/pypy/module/_socket/interp_socket.py#L679-L680

'no-self-use' was moved to an optional extension, i.e. it's now
disabled by default, so no need to expicitly disable it

bad-functions is an option recognized by the bad-builtin
checker, which used to be enabled by default but is now
an optional plugin:
https://pylint.pycqa.org/en/v2.15.2/user_guide/checkers/extensions.html#deprecated-builtins-checker-documentation

The version range in setup.py's greenlet_requires variable is
>=1.1.0, <2.0, but in CI, the upper limit was not specified.

greenlet v2.0.0a1 is out now, and since its PyGreenlet's
data structure is not backwards compatible, geven'ts build
was failing.

Improved documentation of link() and rawlink()