Per-entity load tracking

Not that I know too much about this, but now that I/O would get factored into load, I would think that things like "make --load-average 8" should work better in the face of heavy I/O. The scheduler should also be able to schedule tasks such that the heavy I/O processes aren't run together. I can see some logic behind using the CPU scheduler to help alleviate I/O slowness since the kernel can't do much about I/O speeds, but it can decide when the processes which have a history of making costly I/O get to make more. Memory pressure issues don't make as much sense in the scheduler, but the code covered in the vmpressure_fd article looks promising.