First batch model run

[aappling-usgs]

• made a more complete list of tasks using a merge of the intended forecast date range and those reference dates that are actually available (was an arbitrary set of days before)
• fixed a bug in a column name reference
• changed the 3_forecast_taskmaster.yml recipe for loop_tasks to run all of the tasks rather than just 3
• ran models for 12 hours overnight - that got us about halfway there. @ldecicco-USGS is helping figure out why the WRTDS models ran slower than expected, though 24 hours for the whole run wouldn't destroy us anyway
• gathered the predictions that were made into .rds files and posted them to Drive.

I think anybody can now play with these partial results by calling

sc_retrieve('3_forecast/out/preds_wrtds.rds.ind','3_forecast.yml')
preds_wrtds <- readRDS('3_forecast/out/preds_wrtds.rds')
sc_retrieve('3_forecast/out/preds_loadest.rds.ind','3_forecast.yml')
preds_loadest <- readRDS('3_forecast/out/preds_loadest.rds')