feat: replace Web Speech API with Kokoro TTS for high-quality voices

Adds kokoro-js (82M model, 28 distinct voices) as primary TTS engine with
automatic fallback to speechSynthesis while model loads. All 60+ voice
profiles mapped to real Kokoro voices. Common fight phrases pre-cached
for instant playback.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

frontend/vite.config.ts

@@ -28,6 +28,7 @@ export default defineConfig({
       },
       workbox: {
         globPatterns: ['**/*.{js,css,html,svg,png,woff,woff2}'],
+        maximumFileSizeToCacheInBytes: 5 * 1024 * 1024, // 5MB — kokoro TTS chunk is ~2.2MB
         cleanupOutdatedCaches: true,
         skipWaiting: true,
         clientsClaim: true,