This is part 2 of the serialization cleanup. In this change, we now have

matching structures for serialized comments in Python (defined in the

Review UI files) and JavaScript (defined in the commentData.ts file).

These are fully documented and include cross-references to ensure that

we keep them in sync.
This also makes the serialization stage within the Review UIs more

consistent. One of the giant warts we've had is that the basic ReviewUI

implementation would serialize the comments as a list of dicts, and then

our subclasses would serialize them as a dict, where the key was some

logical grouping (such as x/y/width/height for region comments, or

line/span for text comments) and the value was a list of serialized

comments for that region. We now always serialize to the dict form, and

the base ReviewUI will just use indices for the key and single-value

lists for the values.
A new method has been added to the base ReviewUI to facilitate this, and

just filters and yields the comments. Subclasses can then use this and

sort them however they like. Using this new method fixes a bug in the

TextBasedReviewUI where we weren't doing filtering, and draft comments

from other users could show up when they shouldn't.

Ran js-tests.