[functorch] add batching rule for fill_.Tensor (#84015)
I think this is what the theseus folks ran into, but will confirm with
them later.
Test Plan:
- new manual test; the OpInfo for fill_ isn't sufficient and it is
difficult to modify
Pull Request resolved: https://github.com/pytorch/pytorch/pull/84015
Approved by: https://github.com/Chillee
