Fix 'ResourceWarning's about unclosed 'anyio.streams.memory.MemoryObjectReceiveStream'

[nikitagashkov]

Fixes #2772.

Summary

This PR addresses the cleanup of memory object streams, particularly focusing on ensuring that streams are properly closed to prevent resource warnings:

1. Close recv_stream in BaseHTTPMiddleware upon exception within the route to prevent dangling streams.

Checklist

• I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I've updated the documentation accordingly.

↑ Looks like no changes in docs needed, this is just a bugfix.

[Kludex]

Well, yeah, that's why the pytest.mark.xfail is there... 🤔

[nikitagashkov]

xfail complicates testing for a warning, which is why I had to make adjustments.

An AssertionError within the middleware triggers the same ResourceWarning (refer to issue #2778). However, since the warning is emitted only after the stream is deleted, it appears some time later, specifically after an explicit gc.collect() in test_anyio_streams_cleanup_exc_in_route. This causes the test to fail due to a leftover stream
from another test.

Perhaps there's no need for this test at all, as it is currently quite brittle. 🤷‍♂️ It might be better to stop ignoring the warning from anyio once all streams are properly closed.

[Kludex]

I'm very confused about the comments in this PR. I'll review the other PR.

Adding tons of comments in unrelated code is not helpful.

[nikitagashkov]

I'm very confused about the comments in this PR. I'll review the other PR.

Adding tons of comments in unrelated code is not helpful.

I believe comments are useful for highlighting certain parts of the code, but I have removed them as requested.

Additionally, I made the patchset smaller by removing fixes for TestClient warnings and un-ignore of anyio's ResourceWarning, since the original discussion focused on a specific issue. It's probably better to keep these changes separate as well.

[Kludex]

Let's continue on the other PR.