`difference` and `differenceWith` could be much faster

[sjakobi]

Similarly to intersection and friends (#225), the current implementations of difference[With] leave a lot of performance on the table.

It should be relatively easy to optimize the insertions into the result map with the suggestions from #225 (comment).

Additional gains should be possible by traversing the two trees directly.

unordered-containers/Data/HashMap/Internal.hs

Lines 1748 to 1768 in f59bb26

	-- | /O(n*log m)/ Difference of two maps. Return elements of the first map
	-- not existing in the second.
	difference :: (Eq k, Hashable k) => HashMap k v -> HashMap k w -> HashMap k v
	difference a b = foldlWithKey' go empty a
	where
	go m k v = case lookup k b of
	Nothing -> insert k v m
	_ -> m
	{-# INLINABLE difference #-}
	-- | /O(n*log m)/ Difference with a combining function. When two equal keys are
	-- encountered, the combining function is applied to the values of these keys.
	-- If it returns 'Nothing', the element is discarded (proper set difference). If
	-- it returns (@'Just' y@), the element is updated with a new value @y@.
	differenceWith :: (Eq k, Hashable k) => (v -> w -> Maybe v) -> HashMap k v -> HashMap k w -> HashMap k v
	differenceWith f a b = foldlWithKey' go empty a
	where
	go m k v = case lookup k b of
	Nothing -> insert k v m
	Just w -> maybe m (\y -> insert k y m) (f v w)
	{-# INLINABLE differenceWith #-}

[sjakobi]

It would be good to avoid recomputing the hashes at least.