feat: echte Pair-Audios statt TTS, Klick auf Satz spielt Audio (preloaded)

- usePairAudio-Hook: preload beim Kachel-Mount, nur eine Stimme gleichzeitig
- PairSentenceCard: audio_url statt speechSynthesis (TTS nur noch Fallback)
- PairYesNoCard/PairWordCard: Frage klickbar + Speaker-Button (vorher kein Audio)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/hooks/usePairAudio.js

@@ -0,0 +1,42 @@
+import { useEffect, useRef, useState } from 'react'
+
+// Nur eine Stimme gleichzeitig im Feed
+let currentAudio = null
+
+export default function usePairAudio(url) {
+  const audioRef = useRef(null)
+  const [playing, setPlaying] = useState(false)
+
+  useEffect(() => {
+    if (!url) { audioRef.current = null; return }
+    const audio = new Audio(url)
+    audio.preload = 'auto'
+    const onPlay  = () => setPlaying(true)
+    const onStop  = () => setPlaying(false)
+    audio.addEventListener('play', onPlay)
+    audio.addEventListener('pause', onStop)
+    audio.addEventListener('ended', onStop)
+    audioRef.current = audio
+    return () => {
+      audio.pause()
+      audio.removeEventListener('play', onPlay)
+      audio.removeEventListener('pause', onStop)
+      audio.removeEventListener('ended', onStop)
+      if (currentAudio === audio) currentAudio = null
+      audio.src = ''
+      audioRef.current = null
+    }
+  }, [url])
+
+  function play() {
+    const audio = audioRef.current
+    if (!audio) return false
+    if (currentAudio && currentAudio !== audio) currentAudio.pause()
+    currentAudio = audio
+    audio.currentTime = 0
+    audio.play().catch(() => {})
+    return true
+  }
+
+  return { play, playing }
+}