Categorical matrix optimizations

[ElizabethSantorellaQC]

Replaces Quantco/glum#246
PR: Quantco/glum#253

• Add functionality for F-ordered categorical x dense sandwiches
• Keep speeding up hotspots (currently _check_csc and categorical own-sandwich)
• Extend our data setup (in problems.py) to allow for matrices with mixed dense, sparse, and categorical parts. Currently, it just allows for etiher dense/sparse or dense/categorical.
• Experiment with making small categoricals part of the sparse or dense blocks. Currently, sandwich operations including small categoricals are relatively slow.
• Continue to speed up slow spots, like categorical sandwiched with dense
• Cythonize row and column limiting for categorical sandwiches (currently just have rows)
• Cook up an example where we expect a large performance gain from categorical
• Allow for indices to have various int dtypes
• Add a test for _vec_plus_matvec

[tbenthompson]

Closing in favor of #42