Add nvdaController_isSpeaking to the controller client

[ethindp]

Is your feature request related to a problem? Please describe.

Right now, NVDA is not able to communicate to consumers of the controller client whether it's speaking or not. Applications that wish to know this must therefore settle for screen reader calibration, which involves printing a long string and waiting for the user to press a key (or otherwise signal to the app) to indicate when the screen reader is done. This is extremely error prone, highly inaccurate and unnecessary when NVDA has this capability already.

Describe the solution you'd like

My solution is to add nvdaController3 as an MSRPC interface, exposing isSpeaking([out] boolean* speaking). That would define "speaking" as the synthesizer producing audio at the present time and not as speech as queued (since speech being queued is a different question to pose, and does not intuitively map to what a controller client consumer would expect). PR #20179 performs the necessary work to implement this interface.

Describe alternatives you've considered

The alternative is screen reader calibration, or just not doing anything. This is problematic in things such as RPGs or turn-based games where we need the game to pause and wait but only until the screen reader has finished speaking. Another alternative is to define "speaking" as "speech is queued," but this is problematic because the time between a certain phrase being queued and being spoken are variable and unknown.

Additional context

This functionality would be exposed via the Prism library.

[LeonarddeR]

Have you consideered sending speech to NVDA using SSML? That way you can add SSML marks in your speech and in your library, provide an onSsmlMarkReached callback that gets notifications for marks reached in your speech.