Allow for proper training on new data in pypi

[xmedr]

Overview

In order to get the package to actually train the model on our multi word state data, we needed to append that data to the end of the canonical data. This branch does that and adjusts the instructions in the README to reflect this.

• Connects https://github.com/datamade/parserator.datamade.us/issues/189

Notes

After this comes in I'll push a new version number tag to main to finish the deployment.

Testing

• After pulling down this branch, open a virtual environment and install this version of usaddress
  • pip install -e ."[dev]"
  • It may be useful to uninstall any previous version in your virtual env beforehand, if one exists
• Train the model on just the labeled.xml file
  • parserator train training/labeled.xml usaddress
• Open a python shell and try to parse some addresses with two word state names
  • I used some samples from Zillow

>>> import usaddress
>>> usaddress.parse("6774 68th St S, Horace, North Dakota 58047")
[('6774', 'AddressNumber'), ('68th', 'StreetName'), ('St', 'StreetNamePostType'), ('S,', 'StreetNamePostDirectional'), ('Horace,', 'PlaceName'), ('North', 'StateName'), ('Dakota', 'StateName'), ('58047', 'ZipCode')]