[Accessibility] TTS API interface description + reqs update

[Hynek Hanke]

> > This implies that defer() and say_deferred() are only applicable when
> > the application has chosen the AUDIO_OUTPUT_RETRIEVE method rather than
> > the AUDIO_OUTPUT_PLAYBACK method, because only then does the
> > application know where in the message the defer occurred.

When the method is AUDIO_OUTPUT_PLAYBACK, the application also knows
where in the message the defer occured. In both methods, the application
is receiving information about index marks and events reached. For the
AUDIO_OUTPUT_PLAYBACK method this information comes in form of
registered callbacks (for library call API), asynchronous messages
(socket API) or via a similar mechanism.

Of course the callback mechanism is not completely accurate. It takes
some time to deliver the notifications. If the application needs
completely accurate information, it must receive audio and deal with it.
However, for many practical purposes, such level of accuracy is not
necessary.

With regards,
Hynek Hanke