and thread_at_kernel_exit() are always called for userland threads at the appropriate
situation (note, I've renamed those from *_atkernel_*).
The timing should be more accurate this way, and the thread::last_time_type field
is no longer needed: all interrupts are now added to the kernel time (where the
time is actually spent).
git-svn-id: file:///srv/svn/repos/haiku/trunk/current@11331 a95241bf-73f2-0310-859d-f6bbb57e9c96