Abort TTS earlier in case of fatal errors

It does not make sense for fatal encoding errors to only be propagated *after* all the TTS threads have finished.

Fatal text-to-speech errors are handled not much better (but they are also less likely to happen).

No file, test, or entry point is identified in the issue. Trace the encoding and text-to-speech error handling through the TTS threads, then verify that fatal errors are propagated and processing stops before all threads finish.

Written by the indexing model from the issue text.