The critical wq allows pushing to remote cores and cores never
steal from the critical wq.

if necessary

The give a ~15% improvement in forkexecbench and makes xv6 a bit more
comparable to Linux.

the exec-cleanup and wait-cleanup handlers run on the right core
(disabled by default).

forkexectree now shares relatively little with
steal_nonexec=0 and wq_steal_others=0.

rather than the core that initially called exec.

wait: push finishproc to the exec_cpuid_ core

Conflicts:
	kernel/exec.cc
	kernel/sysfile.cc