The template tag for generating the HTML for replies tried to do a bit

too much, and was not very efficient. It generated replies for the

review body top/bottom fields and for all types of comments, and because

of this mixture it required a lot of state to be passed in. It then

concatenated to a string for every reply, which is not particularly

efficient in Python.
This has been split into two template tags: One for replies to the body

fields of a review, and another to comments. Each only needs to take in

the parameters that are necessary for the type of reply, simplifying

things greatly. It also more efficiently builds the strings,

future-proofs the resulting HTML by explicitly marking as safe, and

transfers some responsibilities to the reply retrieval functions being

called.
There was also a template context push/pop bug in the old code, where we

pushed and then did an update() (implying a push()), and then only

popped once. This caused variables to remain in the context that

shouldn't have been there. It hasn't caused any problems until now, but

it was noticed during development of another change. That bug has been

eliminated.
Unit tests were added for both template tags.

Unit tests pass.
Tested that replies to body top and bottom, general comments, diff

comments, file attachment comments, and screenshot comments all work.
Tested that draft comments show up, and only for the person who made them.