Refactor analyzer

[sanderploegsma]

This ended up being a large refactor, to hopefully make the code base a little easier to understand. Here's a small overview of the largest changes.

Remove input/output handling in exercise analyzers

Previously the Exercise base class was in charge of parsing the input file and writing the JSON output. This has been broken up into several classes to make individual analyzers more easily testable. It also allows a design where multiple analyzers can be 'stacked'.

Add support for comment types

The interface documentation describes an optional type property for comments that was previously not supported. This can now be set using the CommentType enum.

Add support for collecting tags

Although the tags.json file was being written by the analyzer, it was always empty because the mechanism to add tags was not yet implemented.

Use the OO pattern for comments

Previously the analyzer comments were modeled using enums and constants. This makes it hard to keep track of which comments require parameters, and map them to the new CommentType. So instead, each comment is now modeled in a separate class. This class combines the comment key and type into one, and automatically bakes in the parameters.

Remove JSON from unit tests

The exercise analyzer unit tests now only have to check for the presence of the expected comment(s) instead of having to match the JSON output.