[~jbhateja] Proposed for alternative implementations of fillI and copyI using masked ops.
https://github.com/openjdk/jdk/pull/28639#discussion_r2697649279

There are many options here, for example:
- A single loop that only uses masked ops
- Main loop with non-masked ops, post-loop with masked ops
- pre and post loop with masked loops, pre-loop ensures alignment (only works with native memory segment though)
- Possibly some drain loops with smaller vector sizes of non-masked ops.

This would also serve as a study for better handling of small and large loop iteration count JDK-8344085.