take back in some elements from the reverted patch that still make sense based on code review by Moc, thanks

FS-3176 forget the last patch I forgot there is no way to know the right codec using it the file string way, you will just have to manually specify en.PCMA as before but now it should work

change mod_dptools to use the better method of fetching user xml that does not hang onto the xml root

revert 471bd6df and implement reference counting on the xml roots to achieve the same end result

mod_callcenter: Remove the concept of Caller for Members.  Event Socket event have been changed (CC-Caller.* to CC-Member.*)  Also CC-Caller-UUID is renamed to CC-Member-Session-UUID.
The reason for this is you could actually put people to be call in the queue.  So they are not caller per say.  But they are a member of a queue.

mod_callcenter: Generate per member uuid different from the member session uuid.  Might fix transfer between queue.  More changes are commings

FS-3147 --resolve I finally tracked this down to the actual recordings generated by mod_conference.  This patch delays the recording slightly to allow time for the buffer to fill up, we were riding it so closely that sometimes we would come up short and inject silence into the file to preserve time passing

FS-3176 --comment-only try this patch, native is a special case so use the extension native e.g. en.native

FS-3189 ok, patch added, but have you discovered why you trigger this because its not typical and is the sign of a problem I would assume?

introduce new say_string method of doing say and use it in mod_say_en as an example.  try: eval ${say_string en.gsm en current_date_time pronounced ${strepoch()}} from the cli with this patch.  We can do more to centralize the say things and go back and apply it to other langs, using this method you can set the desired file ext as well which I think is a bounty....