Load `BatchML` into the `power_perceiver.load_prepared_batches` code

[JackKelly]

• [ ] Get it working with satellite, PV & GSP
• [ ] Need to make sure we're delivering the correct forecast time_utc for GSP and PV (I assume the forecast timesteps will be missing from the live batches)
• [ ] Get it working with all the numpy batch processors we're using
• [ ] Create load_prepared_batches.data_sources.nwp
• [ ] To start with, lie to the model, and tell it that the PV and GSP ID is always 1. (Background: The ML model requires the PV system row number and GSP ID as a feature. The model has an embedding for these two IDs. Ultimately, before showing PP's predictions to users, we'll need to create a look-up-table that maps from the PV and GSP IDs used in production, to the PV row number and GSP ID expected by PP. But, just to get things working quickly, keep things simple and tell the model that the PV ID and GSP ID are always some fixed number, e.g. 1).

[peterdudfield]

The forecast timestamps for GSP, PV and Satellite will not be present

[peterdudfield]

It might be easier to load Batch into power_perceiver.load_prepared_batches as these will be xarray objects