generate shorter pseudo-random IDs with better spread than current time #1122
Description
There are lots of places in the code where unique IDs are generated mostly based on the current time. That's a big ID space but with lots of correlation between IDs generated by the same run. At some scale these will clash and cause non-deterministic trouble that will be hard to debug.
There is plenty of experience in the world using more spread IDs (c.f. UUIDs)
Everywhere time() is used to generate unique IDs should be replaced by a call to a parsl helper function that does this more sensibly.