Describe the bug

The mean_squared_error metric ignores the squared argument if multioutput="raw_values".
sklearn.metrics.mean_squared_error(multioutput="raw_values", squared=False) behaves as sklearn.metrics.mean_squared_error(multioutput="raw_values", squared=True).

Suggested fix:

On this line:
https://github.com/scikit-learn/scikit-learn/blob/b194674c4/sklearn/metrics/_regression.py#L258
replace the following:
return output_errors --> return output_errors if squared else np.sqrt(output_errors)

Steps/Code to Reproduce

The bug can be reproduced with the following code:

import sklearn
print("sklearn version:", sklearn.__version__)
print(
    "raw values, squared == non-squared:     ",
    sklearn.metrics.mean_squared_error(
        [[1]], [[10]], multioutput="raw_values", squared=True
    )
    == sklearn.metrics.mean_squared_error(
        [[1]], [[10]], multioutput="raw_values", squared=False
    ),
)

print(
    "uniform average, squared == non-squared:",
    sklearn.metrics.mean_squared_error(
        [[1]], [[10]], multioutput="uniform_average", squared=True
    )
    == sklearn.metrics.mean_squared_error(
        [[1]], [[10]], multioutput="uniform_average", squared=False
    ),
)

Which returns the following output:

sklearn version: 0.22.1
raw values, squared == non-squared:      [ True]
uniform average, squared == non-squared: False

Versions

System:
python: 3.7.4 (default, Aug 13 2019, 15:17:50) [Clang 4.0.1 (tags/RELEASE_401/final)]
executable: /Users/usr/.miniconda/envs/env/bin/python
machine: Darwin-19.3.0-x86_64-i386-64bit

Python dependencies:
pip: 19.2.3
setuptools: 41.4.0
sklearn: 0.22.1
numpy: 1.17.2
scipy: 1.3.1
Cython: None
pandas: 0.25.1
matplotlib: 3.1.1
joblib: 0.13.2

Built with OpenMP: True