Fix performance regression in clear_modules!

[fingolfin]

See oscar-system/Oscar.jl#4824 for background.

Unfortunately there is a much worse performance regression on master introduced by PR #2676 (ping @Rahban1). Luckily this one has not yet been in a release of Documenter.

It would be great to have this fix in a release ASAP. Perhaps it would make sense to make a branch for a 1.10.2 release with just this fix but not the other changes, so we have time to properly deal with the other regression (other than reverting the PR in question) ?

[mortenpi]

LGTM!

Perhaps it would make sense to make a branch for a 1.10.2 release with just this fix but not the other changes, so we have time to properly deal with the other regression (other than reverting the PR in question) ?

We haven't done those in a while, but we can definitely do backports when needed.

Side note -- do you think we could add Oscar to be part of our upstream tests? I wouldn't have noticed the performance issues, but it's a big, complex document, so it might reveal a regression sometimes.

Documenter.jl/.github/workflows/regression-tests.yml

Lines 47 to 56 in 10c41a5

	test-documentation-build:
	name: doc-build-${{ matrix.package }}
	runs-on: ubuntu-latest
	env:
	PACKAGE: ${{ matrix.package }}
	strategy:
	fail-fast: false
	matrix:
	include:
	- package: 'MathOptInterface'

[fingolfin]

Thank you for the quick release @mortenpi !

We should open an issue on the other performance regression. (I'd do it now but need to run now to get my son to daycare and then get to teaching. I'll try to remember it later.)

I see no reason why we couldn't/ shouldn't add Oscar to those tests.