Smarter solving

[croyzor]

In our journey to close #28 and #29, there are places where we could be doing better. For example, we fail on the following example:

mapVec(X :: *, Y :: *, { X -> Y }, n :: #, Vec(X, n)) -> Vec(Y, n)
mapVec(_, _, _, _, []) = []
mapVec(_, _, f, _, x ,- xs) = f(x) ,- mapVec(!, !, f, !, xs)

because we can't solve the number argument to the recursive call to mapVec. If we help the typechecker by matching on the number as succ(_) then we can work it out, but this shouldn't be necessary! We know that mapVec preserves vector size, so we should be able to work out from the use of cons that the argument needs to be n - 1

[acl-cqc]

#70 fixes the above example but the following two are still failing even in #41 or the latest branch inference-wip/fork2....here is working source with extra succs, but if you remove the succs then they fail:

repeat(X :: *, n :: #, x :: X) -> Vec(X, n)
repeat(_, 0, _) = []
repeat(_, succ(_), x) = x ,- repeat(!, !, x) -- X can be inferred from x but n cannot

mapFirst(X :: *, Y :: *, { X -> Y}, n :: #, Vec(X, n)) -> Vec(Y, n)
mapFirst(_, _, _, _, []) = []
mapFirst(_, _, f, succ(_), x ,- _) = repeat(!, !, f(x)) -- first ! (X) is second _ (Y)