Performance cleanup. Greatly reduce the number of %fs prefixed globaldata

Performance cleanup.  Greatly reduce the number of %fs prefixed globaldata
lookups by taking advantage of the fact that DragonFly does not preemptively
switch threads to other cpus to cache the globaldata pointer, to pass the
current thread to certain inlines rather then force them to reload it from
%fs, and to use td->td_gd to access the globaldata pointer when td is
available rather then mycpu.

This coupled with the curthread optimization done in a prior commit improved
syscall overheads by 60-100ns (about 7%).