docx writer: avoid extra copy of abstractNum and num elements...

...in numbering.xml.  This caused pandoc-produced docx files to
be uneditable using Word Online.

The problem was that recent versions of reference.docx include
samples of various kinds of text, including lists.  The
numering elements for these were getting copied over to
the new docx, where they clashed with the autogenerated
elements produced by pandoc.  This didn't confuse Desktop
Word, but it did confuse Word Online.

Closes #5358.

src/Text/Pandoc/Writers/Docx.hs

@@ -465,7 +465,15 @@ writeDocx opts doc@(Pandoc meta _) = do
   let numpath = "word/numbering.xml"
   numbering <- parseXml refArchive distArchive numpath
   newNumElts <- mkNumbering (stLists st)
-  let allElts = onlyElems (elContent numbering) ++ newNumElts
+  let pandocAdded e =
+       case findAttrBy ((== "abstractNumId") . qName) e >>= safeRead of
+         Just numid -> numid >= (990 :: Int)
+         Nothing    ->
+           case findAttrBy ((== "numId") . qName) e >>= safeRead of
+             Just numid -> numid >= (1000 :: Int)
+             Nothing    -> False
+  let oldElts = filter (not . pandocAdded) $ onlyElems (elContent numbering)
+  let allElts = oldElts ++ newNumElts
   let numEntry = toEntry numpath epochtime $ renderXml numbering{ elContent =
                        -- we want all the abstractNums first, then the nums,
                        -- otherwise things break: