Implement Tensor.to batching rule (#43206)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/43206
The batching rule is the same as the unary pointwise batching rules:
given a BatchedTensor, we unwrap it, call Tensor.to, and then re-wrap
it.
Test Plan: - `pytest test/test_vmap.py -v -k`
Reviewed By: ezyang
Differential Revision: D23189053
Pulled By: zou3519
fbshipit-source-id: 51b4e41b1cd34bd082082ec4fff3c643002edbaf
