Repeatable model run

We need to ensure we can replicate a model run and know which input files were used to train the model (as these may have changed, but the file name remained the same). When running `parallel-create-projections-and-save` it should also output the input files used to train the model and should include the same prefix and output directory as the parquet files