Fix: MLPRegressor tests

[eddiebergman]

Seems there's been more numerical instability with MLPRegressor. I did a pass over any new source of nondeterminism but could find none.

I've updated the values and started documenting new values that appear while still allowing tests to pass. It seems iterative_fit may be the culprit.

Failing tests: https://github.com/automl/auto-sklearn/runs/4778346386?check_suite_focus=true

[codecov]

Codecov Report

Merging #1367 (8b81345) into development (b01c732) will increase coverage by 0.06%.
The diff coverage is n/a.

❗ Current head 8b81345 differs from pull request most recent head 4a891b6. Consider uploading reports for the commit 4a891b6 to get more accurate results

@@               Coverage Diff               @@
##           development    #1367      +/-   ##
===============================================
+ Coverage        87.97%   88.03%   +0.06%     
===============================================
  Files              140      140              
  Lines            10993    10993              
===============================================
+ Hits              9671     9678       +7     
+ Misses            1322     1315       -7     

Impacted Files	Coverage Δ
autosklearn/evaluation/abstract_evaluator.py	92.27% <0.00%> (-0.78%)	⬇️
..._preprocessing/select_percentile_classification.py	91.52% <0.00%> (+1.69%)	⬆️
autosklearn/pipeline/components/regression/mlp.py	95.31% <0.00%> (+3.12%)	⬆️
...eline/components/feature_preprocessing/fast_ica.py	97.82% <0.00%> (+6.52%)	⬆️
...ponents/feature_preprocessing/select_percentile.py	92.30% <0.00%> (+7.69%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b01c732...4a891b6. Read the comment docs.