Sentinel-2 processing baseline selection

Currently we're using the STAC Collection from https://earth-search.aws.element84.com/v1/collections/sentinel-2-l2a here:

https://github.com/developmentseed/pixelverse/blob/39f7b814f93967391d6b4f1a8a1ab12363315b20/src/pixelverse/download/sentinel2.py#L12-L48

If you look at the search query's returned STAC Item properties, this STAC metadata might show `s2:processing_baseline: "03.01"` for older images, and `s2:processing_baseline: "05.11"` for new images. See details at https://github.com/Element84/earth-search/tree/98868e10793046422efb5c39bc697984ccaa83e6?tab=readme-ov-file#sentinel-2-l1c-and-l2a-sentinel-2-l1c-and-sentinel-2-l2a

If sticking to AWS, we should probably switch to https://earth-search.aws.element84.com/v1/collections/sentinel-2-c1-l2a instead (Collection 1 Level 2A).

Or, if we want to be consistent with [what Tessera was trained on](https://github.com/ucam-eo/tessera/blob/11ff47c7fb46e5d6670c25a70bb176ec73155c9d/tessera_preprocessing/s2_fast_processor.py#L280-L283), we can switch to Planetary Computer on Azure. Other models we may want to include like OlmoEarth are also using Planetary Computer's Sentinel-2 STAC Collection it seems (see [here](https://github.com/allenai/olmoearth_pretrain/blob/b33bc14313db6fe677bdcb762966da64e7f194d3/docs/Dataset-Creation.md?plain=1#L92-L102)).

That said, it's also tricky to know whether Planetary Computer is using Collection 1 or is mixing with the other processing baseline (see https://github.com/microsoft/PlanetaryComputer/discussions/340). So it might pay to propagate the STAC metadata into the output embeddings, so that we know if things need to be reprocessed later due to inconsistencies.

References:
- https://sentinels.copernicus.eu/web/sentinel/-/copernicus-sentinel-2-major-products-upgrade-upcoming
- https://sentinels.copernicus.eu/web/sentinel/-/sentinel-2-collection-1-phase-2-products-availability
- https://planetarycomputer.microsoft.com/dataset/sentinel-2-l2a#Baseline-Change