I've just done two things:

1) Rename the old splnet() to splsoftnet().

2) Implement a new splnet() which blocks network *hardware* interrupts.

Since splimp() must be > spltty(), running network drivers at splimp()
causes the latency of tty interrupts to be higher.  The goal here is
to change the network drivers to use the new splnet(), which can be <
spltty(), rather than splimp().


While doing this, I noticed some errors:

1) In some ports, splnet() already blocked hardware interrupts.  The
pmax port even ran the `le' driver at its incorrect idea of splnet().

2) In other ports, splimp() was < spltty().  This will break with SLIP
or PPP.


I've updated all of the ports in what I think is the correct fashion.
You may want to check your port and ensure that all of the spl levels
are now correct.

(Note: I think splnet() is too high on the MVME68k port, and the SPARC
one is pretty much a WAG.)