Address PyPy flakiness in test_for_leaking_fds

[A5rocks]

Refs #2885

I've run the PyPy test suite a lot and this test in particular has failed.

See https://github.com/A5rocks/trio/actions/runs/6953123771/job/18951033826#step:5:5284 for an example of this happening.

I'm guessing that this is because PyPy GC is weird and isn't guaranteed to clean up previous test objects which might contain something with a file descriptor. I added this here and I didn't get this test to fail again in a bunch of runs, so I think this fixes that.

Now I'm kinda thinking of if it's possible to upstream this in pytest, because this kind of thing is insidious. https://foss.heptapod.net/pypy/pypy/-/commit/5a6a733370542b26e7c4ee756c4fb566523e53d8 introduces a new PyPy GC API that can address this more generally, I believe.

[codecov]

Codecov Report

Merging #2888 (e835f77) into master (9c496fa) will increase coverage by 0.00%.
The diff coverage is 100.00%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2888   +/-   ##
=======================================
  Coverage   99.54%   99.54%           
=======================================
  Files         115      115           
  Lines       17660    17662    +2     
  Branches     3165     3165           
=======================================
+ Hits        17580    17582    +2     
  Misses         52       52           
  Partials       28       28           

Files	Coverage Δ
src/trio/_tests/test_subprocess.py	100.00% <100.00%> (ø)

[jakkdl]

Definitely seems like something that should get fixed somewhere upstream in one way or another, but definitely doesn't hurt to fix in CI for now.

To track the issue one could run the test once before doing gc.collect(), and then rerun it after a collect() on failure - although I'm not sure how to collect data from that.

[A5rocks]

I think the problem is that even without that gc.collect() that tracking method would succeed because the only reason the assertion is failing is because things are getting cleaned up too late. (But after it fails, they're already cleaned up)