[MRG] Add MDD method

[ambroiseodt]

Implement the MDD method from https://arxiv.org/pdf/1904.05801.

Related issue: #254

Core implementation

• Add MDDLoss and MDD in skada/deep/_adversarial.py
• Add test_mdd.py in skada/deep/tests/test_deep_adversarial.py
• Update domain_classifier to deal with multiclass in skada/deep/modules.py

Documentation

• Update accordingly skada/deep/__init__.py
• Update accordingly README.md
• Update accordingly docs in docs/source/all.rst

[codecov]

Codecov Report

Attention: Patch coverage is 95.23810% with 2 lines in your changes missing coverage. Please review.

Project coverage is 96.98%. Comparing base (f4f68b1) to head (2de5598).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #263      +/-   ##
==========================================
- Coverage   97.20%   96.98%   -0.22%     
==========================================
  Files          61       61              
  Lines        6334     6372      +38     
==========================================
+ Hits         6157     6180      +23     
- Misses        177      192      +15     

[tgnassou]

I don't know if we use here the domain Classifier or if we want a DiscClassifier maybe the same that is in the paper

[ambroiseodt]

I changed all the nomenclature "domain" to "disc". In the paper, the disc_classifier must have the same architecture as the task classifier. For the moment, I propose to keep domain_classifier and then do another PR to the adversarial modules from the DomainClassifier (call DomainClassifier or DiscClassifier only in the loss definitions like DANNLoss, MDDLoss instead of the modules like DANN and MDD). It may also be needed to update the DomainAwareNet to change the "domain_classifier" to "disc_classifier".

[antoinedemathelin]

It is almost ok for me
Just two little comments

[antoinedemathelin]

Sounds good to me!