Under load it was possible for the bridging API, and thus ConfBridge, to get
channels that may have hung up stuck in it. This is because handling of state
transitions for a bridged channel within a bridge was not protected and simply
set the new state without regard to the existing state. If the existing state
had been hung up this would get overwritten.

This change adds locking to protect changing of the state and also
takes into consideration the existing state.

ASTERISK-24440 #close
Reported by: Ben Klang

Review: https://reviewboard.asterisk.org/r/4173/


git-svn-id: https://origsvn.digium.com/svn/asterisk/branches/11@428299 65c4cc65-6c06-0410-ace0-fbb531ad65f3