Separate creation of training data and eval data and start eval_state skeleton

[damonbayer]

This PR

• Separates the creation of training data and evaluation data
• Creates eval_state.py, an analogous script to forecast_state.py, to be used for scoring with new data.

To do:

• Verify setup_job.py and forecast_state.py still work.
• Verify post-processing R code works with new file name: data.csv -> eval_data.tsv
• Create new setup_job-like script that runs eval_state.py

For later:

• add scoring to new setup_job-like script that runs eval_state.py

[damonbayer]

This was incorporated into #102