Improvements to BP collective metadata at scale

Created by: williamfgc

Reduced number of Gather/GatherV for indices to 1 Reduced the number of allocations for the metadata buffer Need to test at scale with and without aggregation