possible max_grad_norm implementation error

Clipping is currently applied after the optimizer step. The gradient clipping code controlled by the `max-grad-norm` cli param should be implemented between the backward and step calls.

# Node Prediction Code

https://github.com/awslabs/graphstorm/blob/f3a063669aba254889d4902f6f5c7d512401632e/python/graphstorm/trainer/np_trainer.py#L214-L221

Start in python/graphstorm/trainer/np_trainer.py around lines 214-221 and trace the node prediction training flow from backward through optimizer step. Confirm that max-grad-norm clipping occurs between those calls, then verify the node prediction path still trains correctly with the CLI option enabled.

Written by the indexing model from the issue text.