fit_class_random_forest: process should return training and validation accuracy

[JeroenVerstraelen]

Describe the issue:
The fit_class_random_forest process requires a training parameter which a back-end uses to split the data in a training and validation set. However, there is no description for what should be done with this validation set.

Proposed solution:
Return the training and validation accuracy perhaps as part of the STAC ml-model extension. This allows the user to fine-tune the hyper-parameters based on these values.

[m-mohr]

Yes, I think we should create a PR for the ml-model extension to support this. Once that is done we can clarify this in the process, too. Would you be up to work (with me?) on a ml-model PR?

[jdries]

Dev telco conclusion: drop training/validation split

[m-mohr]

@JeroenVerstraelen @jdries Please review PR #351