Refactor span classes to lower users' learning curve

[frreiss]

This PR makes some changes to our class names to simplify the learning process for new users. The motivation behind this change is the following principle: A user who only needs the functionality of character-based spans should not need to understand token-based spans.

The main changes I've implemented are:

• We now refer to character-based spans as just "spans". The class that used to be called CharSpan is now called Span; CharSpanArray is now called SpanArray; and so on.
• Our dtypes now have names that end in "Dtype", for consistency with how Pandas names its data type objects.
• Instead of returning two columns of per-token spans ("token_span" and "char_span"), all the syntax analysis input functions (SpaCy, CoNLL, and Watson NLU) now return just a column "span" of dtype SpanDtype
• I've updated all the relevant example notebooks to reflect the new nomenclature. We no longer represent each token with both a "char_span" and a "token_span" column.
• Analyze_Model_Outputs.ipynb and Analyze_Text.ipynb still use TokenSpanDtype, but they only use it to store spans that both cover multiple tokens and are constrained to start and end on token boundaries.

[review-notebook-app[bot]]

Check out this pull request on 

Review Jupyter notebook visual diffs & provide feedback on notebooks.

---

Powered by ReviewNB

[frreiss]

CI found a regression. Pushed a fix.