Train model with MERRA2 as NWP input

[jacobbieker]

Following on from getting the data in openclimatefix/satellite-consumer#58 and adding it into datapipes in openclimatefix/ocf_datapipes#308 we should then train a PVNet model to see how the MERRA2 aerosol data affects the performance.

1. Update the configs/models/multimodal.yaml to add another encoder to the nwp_encoders_dict with the key merra2 or cams as we hope to fine-tune it later.
2. Update configs.example/datamodule/configuration/example_configuration.yaml to include the MERRA2 data under the nwp key so that when batches are made, it includes the data
3. Update the config.yaml and configs.example/datamodule/premade_batches.yaml batch_dir to point to the batch directory where the batches have been saved.
4. Run python run.py to run the training!