MM doesn't have any interaction with rfkill at this time, which we probably should have eventually.  This leads to issues with Gobi devices when they get rfkilled and un-killed, they are in low-power state but MM thinks they are still in "on" state because it didn't notice the transition and MM thinks it's the only thing changing power state.

A quick fix would be to double-check the current power state in mm_iface_modem_set_power_state() instead of relying on the cached power state from initialization.

Eventually we probably want plugins to be able to set the power state of the modem if it changes outside of MM, and possibly create a generic power-state handler that listens to the kernel rfkill or something.  This will get tricky though, since there are multiple rfkill "switches" (some hardware and some software) and they interact in non-obvious ways.  Plus, only internal devices actually get rfkilled by the laptop killswitch.  Happy to expand on this if you want more info on the rfkill tragedy since we've been dealing with it in NM for years.