Small fixes

[gabrieletijunaityte]

What does this PR do?

• Remove freezing predictive model twice (now it's done only from setup() method, found in BaseModel class.
• Makes normalisation optional in PredictiveModel and EncoderWrapper to allow control and not normalise features twice.
  • Normalisation is parametrized.
  • Parameters are appended to the trainable modules list.
• Fixed typo in geo_encoder trainable module list generation.
• Added missing mode (train/val/test), appending it to the logging name (for loss and metrics). This is relevant to wandb tracking.
• Encoder wrapper upon setup() call prints the model composition only once now.
• Added aef data cehck (for missing, partial tiles).

Before submitting

• Did you make sure title is self-explanatory and the description concisely explains the PR?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you list all the breaking changes introduced by this pull request?
• Did you test your PR locally with pytest command?

[gabrieletijunaityte]

Sorry, at first I put the wrong target branch.

[robknapen]

Looks good to me. Thanks for finding and fixing these!

[vdplasthijs]

Thanks! Can we change quiet to verbose with logic that if verbose>0 things should be printed, for consistency? Thanks!

[vdplasthijs]

Awesome thanks!