Convert: Fix NemotronH Config Parsing

[anavp-nvidia]

Overview

Fixing NVIDIA-Nemotron-3-Nano-4B-BF16 conversion failure due to config parsing issue (KeyError: '-') with latest versions of transformers (5.5.0 and 5.5.1 at least) .

The fix overrides set_vocab in NemotronHModel with a self-contained implementation that uses trust_remote_code=True. This allows the model's own config class to be loaded, which correctly handles the pattern.

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: Yes, AI tools were used to validate the changes.

[ggerganov]

cc @richarddd @michaelw9999