Adding a featurizer preset with additional matminer featurizer

[gbrunin]

This is related to composition featurizers only.
This adds a preset to the existing ones, with hopefully additional meaningful features.
It also adds the possibility to use only features that are continuous with respect to the composition. Some features are by their nature not continuous, which can lead to unphysical discontinuities when predicting a property as a function of the materials composition.
I have not added tests for this, it should be tested first with matbench to see if it is actually useful or not.

[ml-evs]

Thanks for compiling this @gbrunin, I think this is a useful addition even without fully benchmarking against matbench again. My only general comment, and I'm not convinced myself whether its worth the effort, is whether it makes sense to have the continuous-only vs all features as different presets? This could be very simple in code, i.e. continuous only just inherits from "All" but overrides the lists used in load_featurizer. What do you think?

[gbrunin]

Thanks. I'm not sure about splitting into different presets, because -if I'm not wrong- we would then have 4 presets instead of 2:

• MatminerAll2023Featurizer
• CompositionOnlyMatminerAll2023Featurizer
• ContinuousMatminerAll2023Featurizer
• ContinuousCompositionOnlyMatminerAll2023Featurizer

It could indeed easily be done, but, as you said, not sure it's worth the effort :p let's see what @ppdebreuck thinks.

[ppdebreuck]

Me semble très bien @gbrunin :p . As discussed yesterday, let's keep the number of Featurizer classes to a minimum, and add subsets as params to the constructor. Also, in the future, it would be nice to have a __str__ function that represents nicely the configuration of the featurizer.