This patch does 2 things:
1) Defines a DrwTimer that we use instead of GTimer. This is just a thin
wrapper around g_get_current_time, and it means we can accurately track
typing/idle periods based on real-world wall-clock time, which GTimer is
apparently not intended to do.

2) The typing monitor has some complicated state handling where it transitions
between an IDLE state and a TYPING state. This transition is based on running a
callback once per second, and checking whether any keystrokes have been
recorded since the last time the callback was called. The actual idle *time* is
tracked separately, independently of these two states, but only when we are in
the IDLE *state* was the idle *time* checked to see if we should reset the
break timer. This leads to a race condition -- if we suspend while in the
TYPING state, then eventually we will wake up, notice that no key press has
happened in the last second, *reset the idle timer*, transition to the IDLE
state, and then check the amount of time on the idle timer. We will thus never
notice the amount of time that the computer was suspended.

I considered making the IDLE/TYPING transition code smarter, but it turns out
the desired behavior for the two states is entirely identical anyway, so rather
than adding more complexity to this pointless code, I just diked it out and
replaced them both by a single state called RUNNING.

Closes bug #430797.