Unroll cloneSmallArray when possible

Motivation

Currently, cloneSmallArray# never unrolls the memcpy:

{-# language MagicHash #-}
module CloneFive
  ( cloneFive
  ) where
import GHC.Exts
cloneFive :: SmallArray# a -> SmallArray# a
cloneFive x = cloneSmallArray# x 42# 5#

With -O2, we get this cmm:

     {offset
       cCr: // global
           Hp = Hp + 56;
           if (Hp > HpLim) (likely: False) goto cCv; else goto cCu;
       cCv: // global
           HpAlloc = 56;
           R2 = R2;
           R1 = CloneFive.cloneFive_closure;
           call (stg_gc_fun)(R2, R1) args: 8, res: 0, upd: 8;
       cCu: // global
           I64[Hp - 48] = stg_SMALL_MUT_ARR_PTRS_FROZEN_CLEAN_info;
           I64[Hp - 40] = 5;
           _cCl::I64 = Hp - 48;
           call MO_Memcpy 8(_cCl::I64 + 16, R2 + 352, 40);
           R1 = _cCl::I64;
           call (P64[Sp])(R1) args: 8, res: 0, upd: 8;
     }

Proposal

Unroll the memcpy when the length is statically known. This is already done for memset in several places.