import assert_array_almost_equal from numpy instead of sklearn

[bollwyvl]

What does this PR do?

Avoids the (now private) sklearn.utils._testing import assert_array_almost_equal, just pulling it from numpy (which it's always done)

Where should the reviewer start?

Run tests against sklearn >=0.22

How should this PR be tested?

Run tests against sklearn >=0.22

Any background context you want to provide?

Building downstream on conda-forge/tpot-feedstock#33

What are the relevant issues?

n/a (sry, didn't look)

Screenshots (if appropriate)

n/a

Questions:

• Do the docs need to be updated?
  • no
• Does this PR add new (Python) dependencies?
  • no

[coveralls]

Coverage remained the same at 95.677% when pulling f076653 on bollwyvl:patch-1 into 6448bdb on EpistasisLab:master.

[weixuanfu]

Thank you for this PR. I will merge it to dev branch first.