Increasing max_iter to 1000 for LogisticRegression used in PrejudiceRemover

[mfeffer]

The PrejudiceRemover inprocessing bias mitigation algorithm uses LogisticRegression, and by default, max_iter for LogisticRegression from sklearn is 100. However, in practice, more iterations are occasionally needed to fit PrejudiceRemover to large datasets, some of which are common benchmarks in the fairness literature.

With this in mind, this PR increases the max_iter threshold to 1000 from 100 in order to help PrejudiceRemover properly converge on large datasets.

[nrkarthikeyan]

@mfeffer - please verify the DCO error and sign off.

Also, can we make this a user-configurable option, so that the default is 100 but can be changed to 1000 or more by users who want them? It will make it easier for users who have smaller datasets to run things quickly.

[hoffmansc]

I'm fine with this for now. max_iter is just a budget -- it may use less if possible.

Making this user-configurable will probably have to wait until we rewrite the code for the sklearn-compatible port (#152)