the objection of this patch is to improve the behaviour of the kernel
when reaching a all physical memory used situation.

removed swapfull() checks from fault.c and segment.c as the pagers oom
handler will deal with the situation.  the killbig way of recovering
is more deterministic as the user can prevent it from killing say the
local fileserver or the window system by changing the permissions in
/proc/n.  with swapfull, any process that runs will randomly fail and
if its your local fileserver, the machine is dead.  also, the system will
now behave the same with or without swap file enabled.

changed pio() in port/fault.c:
	only access the s->image while we hold s->lk
	simplified code so demand load and swap in becomes the same
	code path

added ulong gen field to struct Page in port/portfns.h.

pager uses a generation counter (genclock) that it increases after
each walk through the process table.  pages that got referenced/
accessed get the current genclock value assigned.  the pager can now
determine the age of a page by calculating the difference between the
genclock and page->gen.  on each genclock increment, all unreferenced
pages get older by one generation.  on each walk, the pager calculates
the average age of pages wich is used in the next generation to decide
if a page should be swapped or left alone.  if a page is younger or
equal the average age of the pages from the last generation walk, it
is not swapped out.

sort the io transaction list of the pager before processing.

when the pager detects a swapfull, it will invoke the oom path and
call killbig().