Add classes for better dummies encoding, flags for missing data, and missing data replacement

[brettasmi]

I have added the DummiesEncoder, MissingIndicator, and NanReplacer classes to dftransformers.py.

They seem to work well at this point and be robust to np.nan of different dtypes (which I spent yesterday tearing my hair out over). This code was a lot cleaner until I had to add lots of special nan handling. There are still some issues to work out and features to add including:

• code handlers for np.nan in custom DummiesEncoder levels dict
• deal with np.inf and any other special values
• add ability to handle custom missing data values to MissingIndicator and NanReplacer, e.g. -9
• rename NanReplacer to something more appropriate once the above bullet is completed