Update objective computation: multiply quadratic biases with (1-alpha)

[aivarsoo]

The current implementation is not consistent with the documentation, which states that:

   """
    ``alpha=1`` places the maximum weight on the quality of the features,
    and therefore will be equivalent to using
    :class:`sklearn.feature_selection.SelectKBest`.
   """

Closes #17

To make the docs consistent with the code, we change the code and multiply the off-diagonal terms of the objective matrix (or quadratic biases) with 1-alpha

[arcondello]

Looks like the CI failures are not related to this PR. I will fix in a followup.

[arcondello]

It would also be good to have a test that confirms that at the extreme ends of alpha we get what we expect.

[arcondello]

Rather than saving a separate target_correlations array, why not just

Suggested change

	np.multiply(correlations, (1 - alpha), out=correlations)
	np.multiply(correlations[:-1, :-1], (1 - alpha), out=correlations)