Upload trained model to dev bucket

[tjhunter]

First step towards centralized logging: if credentials are provided and if the model is trained, upload it to the dev bucket.

The format should be:

/{run_id}/weathergen_atmo/final_model.pt

Secrets will be loaded from the config file and/or environment variables. I think it is okay to store the AWS access key in the config for the time being.

[clessig]

I don't think we need weathergen_atmo. The run_id should be unique and if we share the same format with land then it can be determined by the id (we could discuss if we choose sub-ranges for land and atmosphere). I would suggest

/{run_id}/checkpoints/final_model.pt

to be flexible for the potential case that we want to stores many more artefacts.