This makes the distribution of pref/min width and the distribution of
percentage width from column-spanning cells more consistent with each
other. Having them inconsistent means that a column-spanning cell that
has both percentage width and minimum intrinsic width might distribute
those widths among its columns differently, which can then lead to the
percentages unnecessarily increasing the table's preferred intrinsic
width.
(It *might* also be worth fixing the FLEX_ALL_LARGE case in
DistributeWidthToColumns to also check for originating cells, since it's
now the only case that doesn't, except that we can't guarantee that
there are columns with originating cells, so it would really require
splitting that case, which is extra complexity that doesn't seem worth
it for such a rarely-hit and non-interoperable case.)
--HG--
extra : transplant_source : %80%C9z%F4%EF%18%E1y%BB%B3%E57%0C%3A%0B%92%2C%B8%83%AC
