Move splitGen to mwc-randon

We have following function in statistics:

``` .haskell
splitGen :: Int -> GenIO -> IO [GenIO]
splitGen n gen
| n <= 0 = return []
| otherwise =
fmap (gen:) . replicateM (n-1) $
initialize =<< (uniformVector gen 256 :: IO (U.Vector Word32))
```

I thing it makes sense to move it to `mwc-random` and generalize to `PrimMonad`. We however need to understand consequences of such splitting algorithm

Start by locating splitGen in the statistics component and reading its current callers and behavior. Then review the mwc-random generator API and assess the consequences of generalizing the split to PrimMonad; done means an agreed algorithm, moved functionality, and coverage for its observable behavior.

Written by the indexing model from the issue text.