preprocessing of era5 sst data format changed for fv3net reservoir computing

[paulaharder]

I changed era_process.py to

• have weekly mean quantities instead of daily
• take start and end dates as arguments to create a dataset for a specific time period
• not have a time dim issue that requires decode_times=False when loading with xarray
• have separate files for different tiles
• enable separate files for train/val/test
• include all given variables (sst,t2m,u10,v10,sst_tend) in one file
• shift t2m,u10,v10
• have a new field "sst_tendency", the sst tendency