feat: streaming audio endpoint with MinIO write-through cache

Add GET /api/audio-stream/{slug}/{n}?voice= that streams MP3 audio to the
client as TTS generates it, while simultaneously uploading to MinIO. On
subsequent requests the endpoint redirects to the presigned MinIO URL,
skipping generation entirely.

- PocketTTS: StreamAudioMP3 pipes live WAV response body through ffmpeg
  (streaming transcode — no full-buffer wait)
- Kokoro: StreamAudioMP3 uses stream:true mode, returning MP3 frames
  directly without the two-step download-link flow
- AudioStore: PutAudioStream added for multipart MinIO upload from reader
- WriteTimeout bumped 60s → 15min to accommodate full-chapter streams
- X-Accel-Buffering: no header disables Caddy/nginx response buffering

backend/internal/kokoro/client.go

@@ -21,6 +21,12 @@ type Client interface {
 	// GenerateAudio synthesises text using voice and returns raw MP3 bytes.
 	GenerateAudio(ctx context.Context, text, voice string) ([]byte, error)
 
+	// StreamAudioMP3 synthesises text and returns an io.ReadCloser that streams
+	// MP3-encoded audio incrementally. Uses the kokoro-fastapi streaming mode
+	// (stream:true), which delivers MP3 frames as they are generated without
+	// waiting for the full output. The caller must always close the ReadCloser.
+	StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error)
+
 	// ListVoices returns the available voice IDs. Falls back to an empty slice
 	// on error — callers should treat an empty list as "service unavailable".
 	ListVoices(ctx context.Context) ([]string, error)
@@ -118,6 +124,49 @@ func (c *httpClient) GenerateAudio(ctx context.Context, text, voice string) ([]b
 	return data, nil
 }
 
+// StreamAudioMP3 calls POST /v1/audio/speech with stream:true and returns an
+// io.ReadCloser that delivers MP3 frames as kokoro generates them.
+// kokoro-fastapi emits raw MP3 bytes when stream mode is enabled — no download
+// redirect; the response body IS the audio stream.
+func (c *httpClient) StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error) {
+	if text == "" {
+		return nil, fmt.Errorf("kokoro: empty text")
+	}
+	if voice == "" {
+		voice = "af_bella"
+	}
+
+	reqBody, err := json.Marshal(map[string]any{
+		"model":           "kokoro",
+		"input":           text,
+		"voice":           voice,
+		"response_format": "mp3",
+		"speed":           1.0,
+		"stream":          true,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("kokoro: marshal stream request: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost,
+		c.baseURL+"/v1/audio/speech", bytes.NewReader(reqBody))
+	if err != nil {
+		return nil, fmt.Errorf("kokoro: build stream request: %w", err)
+	}
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := c.http.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("kokoro: stream request: %w", err)
+	}
+	if resp.StatusCode != http.StatusOK {
+		_, _ = io.Copy(io.Discard, resp.Body)
+		resp.Body.Close()
+		return nil, fmt.Errorf("kokoro: stream returned %d", resp.StatusCode)
+	}
+	return resp.Body, nil
+}
+
 // ListVoices calls GET /v1/audio/voices and returns the list of voice IDs.
 func (c *httpClient) ListVoices(ctx context.Context) ([]string, error) {
 	req, err := http.NewRequestWithContext(ctx, http.MethodGet,