[BUG] CausalLanguageModeling do not mask last input item

Bug description

The clm masking for last item only do not mask last item in input.
It will cause using the embedding of the label instead of mask.

I think following code needs to be fixed.
https://github.com/NVIDIA-Merlin/Transformers4Rec/blob/348c9636399535c566d20e8ebff2b7aa0775f136/transformers4rec/torch/masking.py#L298

Steps/Code to reproduce bug

import torch
from transformers4rec.torch import masking

item_ids = torch.tensor([[1, 2, 0], ])
mask = masking.CausalLanguageModeling(hidden_size=10, train_on_last_item_seq_only=True)
masking_info = mask.compute_masked_targets(item_ids, training=True)
print(masking_info)

MaskingInfo(schema=tensor([[ True,  True, False]]), targets=tensor([[2, 0, 0]]))

Expected behavior

MaskingInfo(schema=tensor([[ True,  False, False]]), targets=tensor([[2, 0, 0]]))

Environment details

• Transformers4Rec version: 23.08.00

Additional context

Start in transformers4rec/torch/masking.py around line 298 and run the provided PyTorch reproduction for CausalLanguageModeling with train_on_last_item_seq_only=True. Compare the current MaskingInfo schema with the expected output; done means the last input item is not masked and the reported result matches the issue.