Cleaned up the nvFuser Python Frontend Batch Norm printing (#88057)
* Removed `define_null_tensor` usage in favor of using optional arguments for binding.
* Re-ordered the non-State arguments for easier printing.
* Added a printing function to include booleans `training` and `channels_last`
* Fixed `define_tensor` to print `is_cpu`
cc @jjsjann123
Pull Request resolved: https://github.com/pytorch/pytorch/pull/88057
Approved by: https://github.com/IvanYashchuk, https://github.com/jjsjann123, https://github.com/mruberry
