[functorch] Added cdist forward/backward batching rules (pytorch/functorch#306)
* WIP on adding cdist batching rules
* Updated cdist forward / backward batch rules
* Fixed code according to the review
- rewrote forward pass reusing BINARY_POINTWISE with an update
- rewrote backward pass + comments
* Restore previous code as cdist issue has been fixed
* Added comment about type promotion for cdist