Optimize runtime `countl_zero`

[andreaslyn]

Branch miss predictions in countl_zero have significant performance impact now that the interpreter is using it for the CLZ instruction. Change the runtime countl_zero function to a branchless implementation.