Reduce memory in classification metrics when average='micro'

[SkafteNicki]

What does this PR do?

Fixes #1279
When average='micro' in multiclass classification we do not need to form the full [n_class, n_class] confmat to do the calculations. This PR introduces a much simpler reduction that scales much better when n_class is a large number.

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[justusschock]

LGTM, can we add a simple test-case asserting that this is actually used properly?

[lucadiliello]

Personally tested and worked flawlessly on MPS, reducing memory usage of the metric from 18GB to a few KB.

[codecov]

Codecov Report

Merging #1286 (5775fb7) into master (afc55c4) will decrease coverage by 37%.
The diff coverage is 100%.

Additional details and impacted files

@@           Coverage Diff            @@
##           master   #1286     +/-   ##
========================================
- Coverage      86%     49%    -37%     
========================================
  Files         190     190             
  Lines       11108   11118     +10     
========================================
- Hits         9607    5490   -4117     
- Misses       1501    5628   +4127     

[SkafteNicki]

LGTM, can we add a simple test-case asserting that this is actually used properly?

Do you want test for memory consumption?
Because tests for correctness is already covered by the tests that are already in place.

[Borda]

Do you want test for memory consumption?

would be great :)