Use .pt2 files to access

to facilitate embeddings generation from more foundation models in a lighter weight fashion, we can explore using .pt2 files instead of brining in model specific inference code.

These .pt2 files could be generated and then live in a common hugging face repo, if they aren't available from the original model provider.

References:
- https://docs.pytorch.org/docs/2.11/user_guide/torch_compiler/export/pt2_archive.html
- https://wherobots.com/solution-gallery/rasterflow-bring-your-own-model/#2-pytorch-2-export-formats
- https://github.com/torchgeo/torchgeo/issues/2732

Start with the linked PyTorch PT2 archive documentation, the Wherobots example, and torchgeo issue #2732 to understand the proposed export format and compatibility constraints. Identify the project entry point for embedding generation and define whether a shared Hugging Face repository of .pt2 files can replace model-specific inference code across the supported foundation models.

Written by the indexing model from the issue text.