Fix optimizer regressions

[andygrove]

Which issue does this PR close?

Closes #3690

Rationale for this change

PR #3662 introduced regressions that were not detected due to a lack of test cases and also due to the fact that we default to skip_failing_rules = true.

What changes are included in this PR?

• Adds tests for the regressions
• Reverts change pre_cast_lit_in_comparison to unwrap_cast_in_comparison #3662

Are there any user-facing changes?

No

[andygrove]

@alamb @liukun4515 This PR is one solution for fixing the recent regressions introduced into the optimizer - by reverting the PR that introduced them - but maybe @liukun4515 can fix these issues in a separate PR, then we won't need to merge this one.

[liukun4515]

@alamb @liukun4515 This PR is one solution for fixing the recent regressions introduced into the optimizer - by reverting the PR that introduced them - but maybe @liukun4515 can fix these issues in a separate PR, then we won't need to merge this one.

I have done the enhancement to fix the bug in #3699

[alamb]

@andygrove can you verify your dask tests work now?

The two tests described above are included now in the optimizer integration test by @liukun4515 : https://github.com/apache/arrow-datafusion/blob/57312284c082c914dd5e1edaa5c1fe3dbe4f222d/datafusion/optimizer/tests/integration-test.rs#L33-L57

[andygrove]

Closed in favor of #3699

#3699 fixes some of the regressions 👍 but there are are more. I will file a new issue shortly once I have a good repro case.