RootCanal HAL: Fixed race condition when tearing down rootcanal HAL

* One race condition exist when incoming_packet_callback_ is being
  reset while an incoming packet is being processed. This is fixed
  by adding a dedicated mutex incoming_packet_callback_mutex_
* Another race condition happen when HciHalHostRootcanal::Stop()
  tries to close the rootcanal FD while incomding_packet_callback
  is trying to read from it

Test: run cert tests multiple times
Bug: 148147248
Change-Id: Ic07374651c5ec38e90e7ddc545ca4fcd4a84cd04