[ENH] add first batch of foundation models into `PTF` for v2.

[PranavBhatP]

Is your feature request related to a problem? Please describe.
There has been a clear demand for integrating foundational models into pytorch-forecasting especially for v2. These foundational models are fundamentally zero shot (and some cases few shot) and their weights are loaded from huggingface. These can be implemented as a child class extension of the v2 BaseModel. This specific class would handled everything from loading the weights, configuring the model and generating prediction. The final prediction API contract would be dictated by the new design for predict in v2.

Some promising candidates for integrating into the library.

1. IBMs TinyTimeMixer
2. Amazon's Chronos.

This issue can serve as a discussion thread for future steps taken in this direction.