Update Linux time IRQ handler to understand the new stolen/blocked cycle counts

Update Linux time IRQ handler to understand the new stolen/blocked cycle counts
exported by Xen. This is based heavily on a patch from Rik van Riel, but
extended to distinguish between idle/blocked cycles and stolen cycles.

There is still stuff todo:
 1. Xen should export the time values in shared memory, to save a hypercall
    on every time interrupt (even though that is only every 10ms, worst case).
 2. As in s390, Xen's cputime_t should be measured at finer granularity than
    jiffies. Nanoseconds would be a better unit.
 3. Break out the internals of the account_steal_time() interface so that we don't
    need to abuse it so wretchedly.

Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Keir Fraser <keir@xensource.com>