[BUG] LLMMaskedCategorical is missing deterministic_sample #4465

Describe the bug

MaskedCategorical exposes both mode and deterministic_sample (the latter returns self.mode). LLMMaskedCategorical implements mode (argmax of the masked logits) but has no deterministic_sample.

Callers that query deterministic_sample (TensorDict CompositeDistribution, ExplorationType.DETERMINISTIC) therefore miss the attribute and fall back on mode with a warning, or raise AttributeError on a direct access.

This is the same gap MaskedCategorical had, but only on the LLM distribution. MaskedCategorical itself is unchanged.

To Reproduce

import torch
from torchrl.modules.distributions import LLMMaskedCategorical

logits = torch.tensor([[[1.0, 3.0, 2.0], [4.0, 0.0, 1.0]]])
mask = torch.ones(1, 2, dtype=torch.bool)
dist = LLMMaskedCategorical(logits=logits, mask=mask)
print(dist.mode)
print(dist.deterministic_sample)

AttributeError: 'LLMMaskedCategorical' object has no attribute 'deterministic_sample'

Expected behavior

dist.deterministic_sample exists and equals dist.mode (argmax of the masked logits), matching MaskedCategorical.

Reason and Possible fixes

Add next to mode in torchrl/modules/distributions/discrete.py:

@property
def deterministic_sample(self):
    return self.mode

Checklist

• I have checked that there is no similar issue in the repo (required)
• I have read the documentation (required)
• I have provided a minimal working example to reproduce the bug (required)