Add vector version of mapAccumL that behaves like the list version

I've implemented a hand rolled version, and another two versions based on a combination of `mapM` and the lazy and strict versions of `State` monad.

https://github.com/haskell-works/hw-prim/pull/38

The benchmarks show that the hand rolled versions run two times faster than the lazy state monad version and 16 times faster than the strict state monad version.

I found the slow performance of the strict monad version most surprising.

I'm aware that the version that using `mapM` might enable fusion, however it is a fair bit slower than a hand rolled version that defeats fusion.

I would love to have a fusion-enabled version that runs as fast as the hand rolled version. Would that be possible?