Better dot rendering of PatternMatch

[acl-cqc]

Similar to #117:

• add edges from PatternMatch to each Box node it mentions
• remove the now-redundant info in the label (OK this also removes detail of what tests are applied - are we happy?)
• Include Boxes owned by a PatternMatch node as siblings of the latter. (The Source+Target and rest will thus be a subcluster)

Also reinstate closures.dot accidentally smashed by #117 :-(

closures.brat f is now:

which I think is an improvement

And g:

Note the "_lhs...._setup" boxes that are not in the hierarchy of the rest of the graph, created here:

brat/brat/Brat/Checker.hs

Lines 738 to 758 in 88e8617

	-- First, we check the patterns on the LHS. This requires some overs,
	-- so we make a box, however this box will be skipped during compilation.
	(sol, match, rhsCty, defs) <- suppressHoles . fmap snd $
	let ?my = my in ("$lhs" -!) $ makeBox (clauseName ++ "_setup") cty $
	\(overs, unders) -> do
	-- Make a problem to solve based on the lhs and the overs
	problem <- argProblems (fst <$> overs) (unWC $ lhs clause) []
	(tests, sol) <- localFC (fcOf (lhs clause)) $ solve my problem
	(sol, defs) :: ([(String, (Src, BinderType m))], [((String, TypeKind), Val Z)]) <- case my of
	Braty -> postProcessSolAndOuts sol unders
	Kerny -> pure (sol, [])
	-- The solution gives us the variables bound by the patterns.
	-- We turn them into a row
	mkArgRo my S0 ((\(n, (src, ty)) -> (NamedPort (toEnd src) n, ty)) <$> sol) >>= \case
	-- Also make a row for the refined outputs (shifted by the pattern environment)
	Some (patEz :* patRo) -> mkArgRo my patEz (first (fmap toEnd) <$> unders) >>= \case
	Some (_ :* outRo) -> do
	let testOuts = snd <$> sol
	let match = TestMatchData my (MatchSequence overs tests testOuts)
	trackM $ "[[[[[[TestMatchData\n" ++ show match ++ "\n]]]]]]"
	pure (sol, match, patRo :->> outRo, fmap (Some . (patEz :*) . abstractEndz patEz) <$> defs)

[acl-cqc]

Prior to the last commit, I also tried putting all the Boxes owned by a PatternMatch in a subcluster - so you'd have subcluster as a sibling of PatternMatch and the Boxes as nieces/nephews rather than siblings, thus there is an extra level of nesting per PatternMatch. Here's f with that approach:

And here's g:

I think the nesting is getting too hard here? But might it be good if there were several sibling PatternMatches?

[acl-cqc]

This should have been removed in #117 too!