Updates for compatibility with newer numpy / PyPI

[gesinecauer]

The following cause errors with the current versions of numpy / PyPI:

• np.float no longer exists, must be replace with float or np.float64
• np.bool no longer exists, must be replace with bool or np.bool_
• 'sklearn' PyPI package is deprecated, must be replaced with 'scikit-learn' for pip commands

This pull request resolves these issues

[gesinecauer]

@NelleV the installation error I'd been having was due to PASTIS being installed before numpy, and I resolved it by letting conda install numpy instead of pip.

Was there a reason you had the environment.yml written such that pip would install most of the dependencies? And is it okay to make this change re numpy installation?

[gesinecauer]

@NelleV looks like the tests are currently failing because it's using the pip-installed version of PASTIS instead of the code from this pull request. The errors are precisely what this pull request was created to address...

How should I best resolve this?

[gesinecauer]

Removing PASTIS from the pip-installed packages in environment.yml resolved the above issue, tests are now run using the pull request code.

Currently, the only failing test is pastis/optimization/tests/test_negative_binomial.py::test_negative_binomial_grad_dense:
assert np.all(grad_dense == 0) fails because grad_dense = array([1.13686838e-13, 0.00000000e+00]).

I'm going to change this line to assert_array_almost_equal(grad_dense, 0).