Suggested improvements to example scripts involving ML model training

[bwohlberg]

The ML model example scripts could be improved by addressing the following two issues:

1. When updating a notebook from an example script using makenotebooks.py, the generated notebook will not include the training run details if the script has previously been run (unless the cached trained model is manually deleted, the need for which is easy to overlook). Some mechanism is needed for signalling to the script/notebook that the cached trained model should be ignored. Perhaps the simplest approach would be to check an environment variable, e.g. SCICO_ML_RETRAIN.
2. The example scripts that use cached training data fail when any of the fixed parameters involved in the data generation are changed. It would be better to include all such parameters in the cache filename (or to include a hash generated from these parameters in the filename) to avoid attempts to load cached data that doesn't match the requirements for a specific run.