[functorch] fix conv_transpose with groups batching rule (#84938)
The original batching rule didn't work in all cases, so this PR fixes it.
Test Plan:
- added new test case to conv_transpose2d's OpInfo. Surprisingly the
other test cases didn't catch the bug.
- Removed some xfails and skips as a result.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84938
Approved by: https://github.com/samdow