onnxruntime

Adding this set_to_none flag to zero_grad to have signature parity with pytorch Adam (#16375)

### Description
torch.optim Adam zero_grad() signature is
zero_grad(set_to_none=True)

https://pytorch.org/docs/stable/generated/torch.optim.Adam.html#torch.optim.Adam.zero_grad

We set this flag in initialization, similar to deepspeed:
https://deepspeed.readthedocs.io/en/latest/optimizers.html#deepspeed.ops.adam.FusedAdam

Adding this flag to have signature parity with pytorch Adam

### Motivation and Context
Easier model integration

Co-authored-by: Jingyan Wang <jingywa@microsoft.com@orttrainingdev7.d32nl1ml4oruzj4qz3bqlggovf.px.internal.cloudapp.net>