As it turns out, smaps/smaps_rollup are fairly slow to read because the
kernel iterates all the memory of a process when this file is accessed.
This caused the update time of all processes to go up from ~30 ms to
more than 200ms.

This changes things to do the reading of smaps_rollup in a separate
thread that doesn't block the main update. This reduces update time to
~50ms, which is noticably better. To accomodate the asynchronous nature,
some additional functionality needed to be added so we can notify
Processes of the asynchronous update.