lapic fixes & xen impact

[Andrew Doran <ad%netbsd.org@localhost>]

I have a fix for the lapic timer to calibrate it properly, which applies to
native x86 only however there is a xen-related impact.

	http://www.netbsd.org/~ad/2020/lapic.diff

With some recent changes for xen it looks like the clocks on x86 were being
started at CPU attach and initialized twice in one case.  However the clocks
on each CPU should be started during the second interrupts-on part of
autoconf (boot cpu) and at the end of cpu hatch (secondary cpus), i.e. when
initclocks() is called.

Other than booting and checking a dom0 and multi-cpu domU with the change is
there anything else specific I should check?

Thanks,
Andrew