In the process, it gains the ability to delete any applied patch (not
just the topmost one, like before), even when deleting patches from
another branch. (However, when deleting patches on another branch, we
obviously can't represent a conflict in the index and worktree, so any
conflicts will make the operation abort.)
One of the t1600 subtests made sure that we couldn't delete
non-topmost patches, and had to be corrected.
Signed-off-by: Karl Hasselström <kha@treskal.com>