WIP: Fix n <= 0 constraint being reported as impossible.

[rowanG077]

The problem is in:

ghc-typelits-natnormalise/src-ghc-9.4/GHC/TypeLits/Normalise.hs

Lines 506 to 524 in 84f500a

	let u' = substsSOP subst (subtractIneq u)
	x' = substsSOP subst x
	y' = substsSOP subst y
	uS = (x',y',b)
	leqsG' | isGiven (ctEvidence ct) = (x',y',b):leqsG
	| otherwise = leqsG
	ineqs = concat [ leqsG
	, map (substLeq subst) leqsG
	, map snd (rights (map fst eqsG))
	]
	tcPluginTrace "unifyNats(ineq) results" (ppr (ct,u,u',ineqs))
	case runWriterT (isNatural u') of
	Just (True,knW) -> do
	evs' <- maybe evs (:evs) <$> evMagic ct knW (subToPred opts leqT k)
	simples subst evs' leqsG' xs eqs'
	Just (False,_) | null k -> return (Impossible (fst eq))
	_ -> do
	let solvedIneq = mapMaybe runWriterT

Given a constraint of a <= 0 then subtractIneq will turn that into -a and isNatural will report it as not a natural. The approach in this PR obviously doesn't work. There is a ton of fallout in the other plugins... So I need to think of something else.

The first commit tried a "proper" fix but there is fallout of tests failing on old GHCs and I tried to compile clash-compiler with it and a ton of breakage occured in ghc-typelits-extra. The second commit added a hack which does not have the problem of fallout. But doesn't work on older GHC due to a <= 0 turning up as a wanted.

Pinging the guru @christiaanb.