The gc module was already imported at module scope in
git/repo/base.py, since f1a82e4 (gitpython-developers#555). Importing the top-level
git module or any submodule of it runs that import statement.
Because the gc module is already imported, reimporting it is fast.

Imports that there is no specific reason to do locally should be at
module scope. Having them local decreased readability, in part
because of how black inserts a black line between them and
gc.collect() calls they are imported to allow.

An alternative to this change would be to remove the preexisting
top-level "import gc" (there is also another one in the test suite)
and replace it with a local import as well. I am unsure if that
would affect performance and, if so, whether the effect would be
good or bad, since the small delay of the import might potentially
be less desirable to an applicaion if it occurs while the work of
the application is already in progress.

If a gc.collect() call runs as a consequence of a finally block or
__del__ method being called during interpreter shutdown, then in
(very) rare cases the variable may have been set to None. But this
does not appear to have been the intent behind making the imports
local. More importantly, a local import should not be expected to
succeed, or the imported module usable, in such a situation.