Fix a couple regressions in the diff viewer.
This fixes two regressions, one new, and one that's been around since

the rewrite.
The new one blocks viewing unpublished diffs. The template code setting

up the diff-related models was assuming the existence of the

latest_diffset object, which didn't exist in this case. The result was

that no value was outputted, leading to invalid JavaScript, which

prevented any diffs from loading.
The second was a visual glitch that resulted in an extra ">" before the

expand links for headers. This was due to the ">" being after the wrong

endif tag.

Posted a diff locally that had both problems. Verified I could view it

before publishing, and that the ">" didn't show up.