We can massage a given selector into a valid predicate register bitmask
and then simply perform a merging move, which eliminates most busywork
around optimizing 256-bit blends.
In the future, once we drop SVE2.1 support in, we can use PMOV to
eliminate the load from memory and related constant management.