[AMD] Trim scheduleGlobalLoadLocalStore to focus on GlobalLoad in prologue (#6409)

The most of `scheduleGlobalLoadLocalStore`'s functionality has been
replaced by the stream pipeliner, so we would like to clean it up. But
moving up the global load ops in the prologue still improves the
performance, especially when there're multiple stages. So we trimmed
this pass to focus on prologue.

Author: knwng