Elect one thread and stick to it

[zasdfgbnm]

In load warp (and prologue), we are doing things like:

for:
  if elect-sync:
    load

That is, for every iteration, we elect a thread, and use that thread to load data.

Looks like it is better to just elect one thread at the beginning, and just use this thread to do all the work. Something like:

if elect-sync:
  for:
    load

See experiment at #3564