This change fixes up a handful of issues related to string types. Some of these are more important than others. For a couple places, we had code deep in the diff rendering which was creating a `SafeString`. Because we have caching at many levels in the diff viewer, these values would end up trying to go through pickle to store the rendered data. When pickle encounters a value that it doesn't understand, it tries to call `str()` on it. With the older Django version, this would correctly convert, but with the modern Django, calling `str()` on a `SafeString` would return the `SafeString` again. This would just end up infinitely recursing inside of the pickle machinery. There are two places where this affects. For the first, the fix is to remove some old `mark_safe` calls. For the second (where we were rendering a template), the solution is to explicitly convert to `str` by using a slice trick. Next, the HTTP digest authentication backend was calling MD5, but passing in an `str` instead of `bytes`. We also had a handful of other places where there was some confusion about bytes/str which was found via static analysis. Finally, a bunch of unit tests were passing in various types of objects (int, Revision) when creating test filediffs. This would end up getting cast appropriately, but was tripping up some tooling. I've changed these to explicitly pass in the strings. Testing Done: - Created a review request with a diff and was able to successfully load the diff viewer page. - Ran unit tests.