The Review UI support had pretty minimal tests, which meant that some

recent issues weren't caught earlier in Python 3.x development. The few

tests we did have for the core ReviewUI and FileAttachmentReviewUI

were more about the sandboxing of some calls, which didn't really help

as much of the sandboxing applied the wrong approach (ignoring faults in

subclasses, resulting in harder-to-diagnose issues, rather than outright

failing).
Sandboxing is important when common code is calling into

extension-provided code, but in these cases we were dealing with

extension-provided code (subclasses of ReviewUI) calling other

extension-provided code (overridden methods of that same ReviewUI).
This removes all the old testing code, adding new more comprehensive and

correct tests for ReviewUI and FileAttachmentReviewUI. These live in

the standard reviewboard.reviews.tests directory, rather than the more

buried reviewboard.reviews.ui.tests (which is no more). The existing

ImageReviewUI (which could still use additional work) has been moved

into reviewboard.reviews.tests as well.
There were a few changes made to ReviewUI as a result of the tests.

The inline parameter (intended for rendering a ReviewUI within

something like the diff viewer, but never fully used in practice) now

does a better job of processing input in a request, and

get_comments_json() no longer tries to mark the string as safe

(leaving that up to the template).

Unit tests pass.