feat(workspace): voice notes via /audio (M3)

- /audio (only in an empty bullet) shows a recording bubble: green round
  button starts, it turns red with a live wave and timer on its left, a
  tap saves the recording as an inline audio player. An audio file can be
  uploaded instead; Esc/✕ restores the empty bullet.
- Chrome's WebM recordings get their length measured, so the player shows
  the total time and seeking works.
- With the own Whisper server "Transkribieren" turns the recording into
  the (searchable) caption.
- Loaded files keep their Blob, as the CSP rightly blocks fetch() on blob:
  URLs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

frontend/src/features/workspace/AudioBlock.tsx

@@ -0,0 +1,136 @@
+import React, { useEffect, useRef, useState } from 'react';
+import { FileAudio, Loader2, Wand2 } from 'lucide-react';
+import { api } from '../../lib/api';
+import { mediaBlob, mediaObjectURL, parseMediaContent } from '../../lib/media';
+import { loadVoiceConfig } from '../../lib/voice';
+
+interface AudioBlockProps {
+  content: string;
+  canEdit: boolean;
+  onCaptionChange: (caption: string) => void;
+}
+
+// Audio block: player plus caption. With the own Whisper server the recording can be turned into text,
+// which becomes the caption (and so searchable).
+export const AudioBlock: React.FC<AudioBlockProps> = ({ content, canEdit, onCaptionChange }) => {
+  const ref = parseMediaContent(content);
+  const mediaId = ref?.id;
+  const [src, setSrc] = useState<string | null>(null);
+  const [failed, setFailed] = useState(false);
+  const [caption, setCaption] = useState(ref?.caption || '');
+  const [canTranscribe, setCanTranscribe] = useState(false);
+  const [transcribing, setTranscribing] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+  const audioRef = useRef<HTMLAudioElement>(null);
+
+  // Recordings from Chrome (WebM) carry no length: jump to the end once so the browser measures it,
+  // otherwise the player shows no total time and seeking does not work
+  const fixDuration = () => {
+    const a = audioRef.current;
+    if (!a || Number.isFinite(a.duration)) return;
+    const reset = () => {
+      a.removeEventListener('timeupdate', reset);
+      a.currentTime = 0;
+    };
+    a.addEventListener('timeupdate', reset);
+    a.currentTime = 1e101;
+  };
+
+  useEffect(() => {
+    setCaption(ref?.caption || '');
+  }, [ref?.caption]);
+
+  useEffect(() => {
+    if (!mediaId) return;
+    let alive = true;
+    setFailed(false);
+    mediaObjectURL(mediaId)
+      .then((u) => alive && setSrc(u))
+      .catch(() => alive && setFailed(true));
+    return () => {
+      alive = false;
+    };
+  }, [mediaId]);
+
+  useEffect(() => {
+    void loadVoiceConfig().then((cfg) => setCanTranscribe(cfg.mode === 'server'));
+  }, []);
+
+  if (!ref) {
+    return <div className="text-xs text-status-err py-1">Audio-Verweis ist beschädigt: {content}</div>;
+  }
+
+  const commitCaption = (value: string) => {
+    if (value.trim() !== ref.caption) onCaptionChange(value.trim());
+  };
+
+  const transcribe = async () => {
+    if (!src) return;
+    setTranscribing(true);
+    setError(null);
+    try {
+      const blob = await mediaBlob(ref.id);
+      const text = await api.transcribe(blob);
+      if (text) {
+        setCaption(text);
+        commitCaption(text);
+      } else {
+        setError('Nichts erkannt');
+      }
+    } catch (err: any) {
+      setError(err?.message || 'Spracherkennung fehlgeschlagen');
+    } finally {
+      setTranscribing(false);
+    }
+  };
+
+  return (
+    <div className="my-1 space-y-1">
+      {failed ? (
+        <div className="flex items-center gap-2 h-12 px-3 rounded-card border border-border bg-panel text-xs text-muted">
+          <FileAudio className="w-4 h-4" /> Aufnahme nicht verfügbar (offline oder gelöscht)
+        </div>
+      ) : src ? (
+        <div className="flex items-center gap-2">
+          <audio ref={audioRef} controls preload="metadata" src={src} onLoadedMetadata={fixDuration} className="w-full max-w-md h-11" />
+          {canEdit && canTranscribe && (
+            <button
+              type="button"
+              onClick={() => void transcribe()}
+              disabled={transcribing}
+              title="In Text umwandeln (eigene Spracherkennung)"
+              className="btn-secondary h-11 sm:h-8 px-2.5 text-xs flex items-center gap-1.5 shrink-0 disabled:opacity-40"
+            >
+              {transcribing ? <Loader2 className="w-3.5 h-3.5 animate-spin" /> : <Wand2 className="w-3.5 h-3.5" />}
+              <span className="hidden sm:inline">{transcribing ? 'Wird erkannt…' : 'Transkribieren'}</span>
+            </button>
+          )}
+        </div>
+      ) : (
+        <div className="flex items-center h-12 px-3 rounded-card border border-border bg-panel w-64">
+          <Loader2 className="w-4 h-4 animate-spin text-muted" />
+        </div>
+      )}
+
+      {canEdit ? (
+        <input
+          type="text"
+          value={caption}
+          onChange={(e) => setCaption(e.target.value)}
+          onBlur={() => commitCaption(caption)}
+          onKeyDown={(e) => {
+            if (e.key === 'Enter') {
+              e.preventDefault();
+              (e.target as HTMLInputElement).blur();
+            }
+          }}
+          placeholder="Beschriftung hinzufügen…"
+          className="w-full bg-transparent border-0 p-0 text-xs text-muted placeholder-faint focus:outline-none focus:ring-0"
+        />
+      ) : (
+        ref.caption && <p className="text-xs text-muted">{ref.caption}</p>
+      )}
+      {error && <p className="text-xs text-status-err">{error}</p>}
+    </div>
+  );
+};