Move Windows sound API to right thread context and make it an object member. Perform init within the right thread context. Add capability to play .wav files in addition via QMultiMedia

This function is called automatically when a QSettings object is destroyed,
so is rarely needed to be called explicitly.

There are still some cases where
a QSettings object is passed-by-reference, so there I left the sync() calls as
it's unclear how permanent those objects are and if they'd call sync() before
application shutdown

Changed to get the current voice properties, then modify then and set new ones.

Changed to get the current voice properties, then modify then and set new ones.

Found via valgrind.

Broke windows compilation.

Add debug output for when voice initialization fails on Windows

Removed or annotated existing commented-out code

Clean up logic for saying emergency text.

Previous code didn't properly check for if speech was enabled.

espeak: provide string length with 0 character, fixes problems with qgc reading out strange values/chars at the end of the sentence

All it does is clutter the terminal and it's not doing anything anyways.

Speech synthesis support can now be disabled (DEFINES+=DISABLE_SPEECH)
Speech synthesis dependencies are now checked before enabling.
When support cannot be enabled, the user is now notified.

Conflicts:
	QGCExternalLibs.pri
	src/GAudioOutput.cc

While testing the previous commit on Windows I found that I could not
compile windows due to include of atlbase.h. This file is no longer
available in SDKs. It is only available in for money versions of VIsual
Studio. Luckily this include was not needed. So I just removed it and
Windows now builds again with just VS Express. Tested all changes on
OSX, Linux and Windows (including speech on windows, which still works).