"Strict" mode for voice selection

There's currently no way to control which voices are allowed as fallback for other voices, so that the result is possibly a mix of voices that don't sound the same.

There is also no way to tell Pipeline that no fallbacks are allowed, or that no "non-exact" fallbacks are allowed. Pipeline always tries exactly one fallback, no more and no less, and it tries a fallback no matter whether it is an "exact" match or not. The user only gets to see a warning in the TTS log when a voice does not match the requested characteristics exactly.

The only thing a user can do at the moment is select preferred voices, which will influence the fallbacks that will be chosen by the system.

No files, tests, or entry points are identified in the issue. Start by locating Pipeline's voice-selection and fallback handling, then define how strict and non-exact fallback options should be exposed and verify that fallback behavior follows those choices.

Written by the indexing model from the issue text.