Add support for IBM Granite 4.0 Tiny Preview

[donpellegrino]

It would be interesting to explore the IBM Granite 4.0 models. Information is available at https://www.ibm.com/new/announcements/ibm-granite-4-0-tiny-preview-sneak-peek and https://huggingface.co/ibm-granite/granite-4.0-tiny-preview. It has been added to the HuggingFace Transformers library with huggingface/transformers#37658. As of the time of this post, conversion to ONNX via HuggingFace Optimum (https://huggingface.co/docs/optimum/main/exporters/onnx/usage_guides/export_a_model) fails with an error that Optimum does not yet support the model.

Relevant Burn documentation for an implementation seems to be https://burn.dev/burn-book/import/pytorch-model.html and https://dev.to/laggui/transitioning-from-pytorch-to-burn-45m.

[donpellegrino]

Config
The HuggingFace Transformers Config specifies "mamba_d_head" as an int with a default of the string "auto." This seems to be a bug in Transformers (https://github.com/huggingface/transformers/blob/d69945e5fced637e77cf7af5e4955cb897bc298c/src/transformers/models/granitemoehybrid/configuration_granitemoehybrid.py#L109C9-L109C21).