Add L-TAE model

[MUYang99]

Add L-TAE model for satellite image time series processing: https://arxiv.org/pdf/2007.00586, https://github.com/VSainteuf/lightweight-temporal-attention-pytorch

[MUYang99]

@microsoft-github-policy-service agree

[isaaccorley]

Thanks for the PR! Looks like the code is missing type hints throughout and needs to be formatted and limited with ruff! But otherwise lgtm.

Are there are any pretrained weights available for L-TAE?

Edit: also need to add the model to hubconf.py and torchgeo/models/api.py

[isaaccorley]

@MUYang99 any thoughts on just scrapping the custom implementations of MHA, SDPA, and positional embeddings with the implementations from the torch libraries?

I think that torch has more optimized versions of MHA and SDPA. I think the LTAE code is just very old so the implementation didn't have these classes available in torch yet.

[adamjstewart]

Can you resolve the merge conflicts with docs/api/models.rst? This will then allow the tests to run.

[isaaccorley]

Just need to format with ruff format && ruff check and fix mypy complaints and this is good to go!

[isaaccorley]

This lgtm @adamjstewart any final nits?