Add more constraints on RankNTypes function args

[DigitalBrains1]

We don't always specify all constraints that could be specified on Nat in our functions. For instance, riseEvery requires an SNat larger than zero, but it also accepts zero, which causes a runtime error. Usually it's not a problem because we can pass a more-constrained argument to a less-constrained function.

But when we're talking about the argument of a higher-order function (using higher-rank type), the reverse is true. We have

smap :: forall k a b. KnownNat k => (forall l. SNat l -> a -> b) -> Vec k a -> Vec k b

(same for dfold, dtfold, vfold). This precludes using functions that restrict l (ell) to be smaller than k, like at, take, drop, select, splitAt, etc. These all require l be smaller than k, and GHC/CλaSH can't prove that this is the case.

This lead to this Google Group thread.

We should add constraints on these function arguments to allow using constrained functions.

[leonschoorl]

Same idea as #1054