This was an oversight when doing thread suspension, and makes sure create_thread and schedule_thread behave the same way in terms of not scheduling on suspended threads.

Proposed Changes

• Add a select_active_pu function to scheduler_base which accepts an initial thread number (this one is tried first, others are tried sequentially, wrapping around when needed), and an optional fallback thread number (used in case locks can't be taken immediately to avoid looping forever; the fallback is used in scheduling_loop to avoid looping for too long when the current pu could do work)
  • None of the above is done if elasticity is disabled