fix(tts): make stepfun preset pick case-stable and per-character

- Hash the lowercased description (matching the case-insensitive scoring)
  so the same archetype text picks the same preset regardless of case.
- Thread the character name through provisionVoice -> stepfunProvision as
  the hash salt, so two characters that share archetype keywords spread
  across the top-N candidate presets instead of collapsing on one voice.

Xiaomi path is unaffected (voicedesign mints a unique clip per call).

app/play/page.tsx

@@ -525,7 +525,7 @@ async function resolveByoVoice(
     return ready;
   }
   if (!speaker.voiceDescription) return null;
-  const p = provisionVoice(cfg, speaker.voiceDescription);
+  const p = provisionVoice(cfg, speaker.voiceDescription, speaker.name);
   cache.set(speaker.name, p);
   try {
     return await p;