Fix/nlinear normalization for multivariate series

[madtoinou]

Fixes #2035.

Summary

• Added slicing of the input tensor x when normalizing (only the target features based on the last value of the target features)

Other Information

• The problem does not occur when shared_weights=True because covariates are not allowed (and normalization is applied to all the dimensions).

[codecov-commenter]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (36a1c09) 93.85% compared to head (4edbaa9) 93.85%.

Files	Patch %	Lines
darts/models/forecasting/nlinear.py	0.00%	1 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2072      +/-   ##
==========================================
- Coverage   93.85%   93.85%   -0.01%     
==========================================
  Files         134      134              
  Lines       13123    13109      -14     
==========================================
- Hits        12317    12303      -14     
  Misses        806      806              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[dennisbader]

Looks great, thanks for fixing this @madtoinou

[felixdivo]

Thanks @madtoinou!