Revert PR #3011 with additional test changes

[rambleraptor]

Rationale for this change

This change reverts #3011 since we've seen a regression.

I've verified this against the reproduction in #3758.

There's one additional test change in this PR, since the test in question did not exist #3011 was merged in and will fail without #3011.

Are these changes tested?

Tests still pass + verified the reproduction

Are there any user-facing changes?

[nssalian]

The PR is doing both a revert and adding tests for the reproduction. Could this cleanly revert first and we could enhance testing in a follow up?

[rambleraptor]

The test change isn't to make it match the reproduction.

This is the cleanest version of the revert that we're able to make given that the PR was merged 6 months ago. The test that I had to alter didn't exist when the original PR was merged in.

[nssalian]

The test change isn't to make it match the reproduction.

This is the cleanest version of the revert that we're able to make given that the PR was merged 6 months ago. The test that I had to alter didn't exist when the original PR was merged in.

Thanks for clarifying. Do you mind adding a note on the PR description so that is clear. The title and the description threw me off.

[rambleraptor]

@nssalian did both. thanks for the punctual review!