kernel - Revamp LWKT thread migration

kernel - Revamp LWKT thread migration

* Rearrange the handling of TDF_RUNNING, making lwkt_switch() responsible
  for it instead of the assembly switch code.  Adjust td->td_switch() to
  return the previously running thread.

  This allows lwkt_switch() to process thread migration between cpus after
  the thread has been completely and utterly switched out, removing the
  need to loop on TDF_RUNNING on the target cpu.

* Fixes lwkt_setcpu_remote livelock failure

* This required major surgery on the core thread switch assembly, testing
  is needed.  I tried to avoid doing this but the livelock problems persisted,
  so the only solution was to remove the need for the loops that were causing
  the livelocks.

* NOTE: The user process scheduler is still using the old giveaway/acquire
	method.  More work is needed here.

Reported-by: "Magliano Andre'" <masterblaster@tiscali.it>