New setting to make TTS timeouts less strict

The rationale of this enhancement is that there can be various reasons why a `synthesize()` call can take longer than expected (and more than the already built-in tolerance factors). E.g., the timeout system is not very well adapter to cloud TTS services (which were implemented later). A timeout is set on the `synthesize()` method, but TTS engines based on a cloud service may do retries of a request when a server is busy, and this all happens within one `synthesize()`. In such cases the timeout does not give much room for something like an exponential backoff strategy.