Retrain base model to account for reprocessed SCL band

[ejm714]

"Improved L2A processing algorithms for scene classification and surface reflectance" have been applied to the historical archive of Sentinel-2 data (Source). This has a couple impacts on our pipeline:

• the cloud threshold depends on the SCL band. changing the number of cloud pixels detected can affect whether that image is a candidate or not
• water pixel identification depends on the SCL band. changing the number of water pixels directly affects the pct_water feature and indirectly affects all the band statistics calculated on the subset of water pixels

While slight changes in pixels included shouldn't have a dramatic affect on accuracy, it would be best to retrain the model with the updated features to ensure consistency between data processing in model training and data processing in prediction.