The diff viewer had a caching bug that people noticed when using

interdiffs. Essentially, updating the diff and viewing its interdiff

resulted in the old diff being displayed.
The reason this happened is that DiffFragmentView.make_etag() was

expecting a DiffSet ID, but this wasn't actually available at this

point. We didn't have DiffSet information until later, after the caching

checks, so we couldn't include any of its identifying information.
Rather than regress the performance of the diff viewer by moving the

database access before the caching checks, I've opted to extend the URL

format to take an interfilediff ID. This gives us all the information

needed to sanely cache fragments without incurring any performance

penalties.

Tested diffs and interdiffs of different types in the diff viewer. Saw that

the content had a new ETag every time a diff was updated, and for every

diff revision range.
Unit tests pass (both Python and JavaScript).