I've noticed that sometimes when our CI server is under high load, we

were getting an intermittent test failure in

ExtensionManagerTests.test_install_media_concurrent_threads.
This lead me to discover that in Django 3.2,

django.core.files.locks.lock() changed its interface to return False

instead of raise IOError. This meant that our use of this has not

actually been using the lock at all. This change updates the extension

manager to check the return value and raise an IOError to restore the

behavior described in the docstring.

Ran unit tests.
Artificially created a high load situation and verified that prior to

  the fix we would hit contention, and after we properly respect the lock.
Grepped through our codebases to ensure there weren't any other

  similar bugs.