Update torch.norm to torch.linalg.norm

[loadams]

Todo:

• Update PR since torch.norm and torch.linalg.norm have different function signatures.
• Check if there are any numeric differences between the functions.
• Determine why there appear to be performance differences (nv-accelerate-v100 most notably: 2245.12s call tests/deepspeed/test_deepspeed.py::DeepSpeedConfigIntegration::test_basic_run)
• Update to torch.linalg.matrixnorm or torch.linalg.vectornorm