We only need to resume the SUSPENDED node when it has !pause-on-idle.
So, set the need_resume flag when we suspend in that case.

Clear the flag in the next round and make sure we set the node to
running again. We don't need to clear any flags when the state change
completes anymore with this change.

This also fixes a case where strawberry would fail to configure the
correct sample rate. 44.1KHz would be used (the rate of the probe
stream) because the scheduler was thinking the node was still being
reconfigured because the reconfigure flag was only cleared when going
to the running state (an alsa sink has !pause_on_idle).

See #2929