look into incorrect diffs produced by patchutil.CombineDiff #222

the algorithm is slightly wrong, or needs to handle an edge-case: it sometimes adds empty lines for lines that are unknown and these are rendered in the final diff. it remains to be seen if CombinePatch is even possible without a git-index, one fix is to request the combined patch from the knotserver for a given revlist and cache that on the AV.