Multiple channels logging in as the same agent can result in dead channels
waiting for a condition signal that will never come because another
channel thread stole it.  A symptom is chan_sip repeatedly generating
warning messages about rescheduling autodestruction of dialogs with an
agent channel owner.

* Made only login_exec() (the app AgentLogin) clear the agent_pvt->chan
pointer to prevent multiple channels from logging in as the same agent.
agent_read(), agent_call(), and agent_set_base_channel() no longer
disconnect the agent channel from the agent_pvt.  This also eliminates the
need to keep checking for agent_pvt->chan being NULL.

* Made agent_hangup() not wake up the AgentLogin agent thread until it is
done.

* Made agent_request() not able to get the agent until he has logged in
and any wrapup time has expired.

* Made agent_request() use ast_hangup() instead of agent_hangup() to
correctly dispose of a channel.

* Removed agent_set_base_channel().  Nobody calls it and it is a bad thing
in general.

* Made only agent_devicestate() determine the current device state of an
agent.  Note: Agent group device states have never been supported.

Review: https://reviewboard.asterisk.org/r/2260/
........

Merged revisions 380364 from http://svn.asterisk.org/svn/asterisk/branches/1.8


git-svn-id: https://origsvn.digium.com/svn/asterisk/branches/11@380384 65c4cc65-6c06-0410-ace0-fbb531ad65f3