Separate functionality in postprocess_state_forecast.R

[damonbayer]

          In the meantime, this should probably be moved to it's own script. In that script we can save each group as a separate `parquet` (as would be done with an equivalent python script) and then load only the necessary parquets here.

Originally posted by @damonbayer in #102 (comment)

This change will result in a workflow be closer to flowchart and make swapping out for a forecasttools-py solution easier.