nvprims native batch norm patch (#88455)
Cherry-picking: https://github.com/csarofeen/pytorch/pull/2104
- [x] Added explicit cast on inputs to nvprims.native_batch_norm. This avoids the explicit cast, which gives us issue on fusion definition.
- [x] add python repro with dynamo
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88455
Approved by: https://github.com/mruberry, https://github.com/IvanYashchuk
