Revert "Apply input sharding to all compatible input tensors"

[alanwaketan]

Reverts #4872

This breaks mnist.

root@t1v-n-307ffe96-w-0:/workspaces/work/pytorch/xla# PJRT_DEVICE=TPU python test/test_train_mp_mnist.py --num_epochs=1
Traceback (most recent call last):
  File "test/test_train_mp_mnist.py", line 2, in <module>
    from torch_xla.experimental import pjrt
  File "/workspaces/work/pytorch/xla/torch_xla/experimental/pjrt.py", line 14, in <module>
    import torch_xla.core.xla_model as xm
  File "/workspaces/work/pytorch/xla/torch_xla/core/xla_model.py", line 11, in <module>
    import torch_xla.experimental.xla_sharding as xs
  File "/workspaces/work/pytorch/xla/torch_xla/experimental/xla_sharding.py", line 10, in <module>
    from torch_xla.experimental.pjrt import requires_pjrt
ImportError: cannot import name 'requires_pjrt' from partially initialized module 'torch_xla.experimental.pjrt' (most likely due to a circular import) (/workspaces/work/pytorch/xla/torch_xla/experimental/pjrt.py)

[JackCaoG]

@jonb377 can you take care of the reland?

[alanwaketan]

Thanks Jack for approving.