Maybe: add {before,after}_io_wait callbacks to instrument interface?

[njsmith]

It's trivial to do and gives useful information; I'm just a bit hesitant because depending on the outcome to #32, we might or might not have a discrete handle_io step in the scheduler loop.

[njsmith]

On further thought, it still makes sense even if we change the scheduler loop -- we just might be doing more calls to check for IO, with 0 timeout. So I added this.