I've just commmitted another slew of changes to the kernel
autoconfiguration and initialization code.  The TC and ioasic
initialization code is cleaned up. On the 3100, baseboard deviecs now
attach to a virtual "ibus" bus, instead of straight to the mainbus.

These are reasonably well tested; the only exception is the 5000/200,
which I no longer have access to.  Almost all of these got into
today's sup; I forgot a couple of changes to ibus devices in
sys/dev/tc, which won't show up until tomorrow's sup.

The big change is preliminary support for the 5100, based on the
"ibus" support.  Simon Burge has been extremely helpful with remote-
debugging the 5100 code.  A 5100 kernel gets as far autoconfiguring
devices, mounting the root filesystem, forking off initialization
processes, and exec'ing /sbin/init.  At that point it hangs.

I don't have access to a 5100 myself, and I'm out of ideas to remotely
debug, so I've committed these changes to let other people keep
working on them, and so that some other pending cleanups can go into
the pmax tree.

There's also support for using the mips3 CPU cycle-counter as a
high-res clock on 5000/150.  The 5000/50 has a baseboard free-running
1Mhz clock, and NTP says the FRC has a accurate crystal than the CPU,
so for now the 5000/50 is using the baseboard counter.  (I don't have
a 5000/260 to test.)