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/pockettts/client.go

@@ -9,6 +9,10 @@
 // so callers receive MP3 bytes — the same format as the kokoro client — and the
 // rest of the pipeline does not need to care which TTS engine was used.
 //
+// StreamAudioMP3 is the streaming variant: it returns an io.ReadCloser that
+// yields MP3-encoded audio incrementally as pocket-tts generates it, without
+// buffering the full output.
+//
 // Predefined voices (pass the bare name as the voice parameter):
 //
 //	alba, marius, javert, jean, fantine, cosette, eponine, azelma,
@@ -50,6 +54,11 @@ type Client interface {
 	// Voice must be one of the predefined pocket-tts voice names.
 	GenerateAudio(ctx context.Context, text, voice string) ([]byte, error)
 
+	// StreamAudioMP3 synthesises text and returns an io.ReadCloser that streams
+	// MP3-encoded audio incrementally via a live ffmpeg transcode pipe.
+	// The caller must always close the returned ReadCloser.
+	StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error)
+
 	// ListVoices returns the available predefined voice names.
 	ListVoices(ctx context.Context) ([]string, error)
 }
@@ -79,14 +88,97 @@ func (c *httpClient) GenerateAudio(ctx context.Context, text, voice string) ([]b
 		voice = "alba"
 	}
 
-	// ── Build multipart form ──────────────────────────────────────────────────
+	resp, err := c.postTTS(ctx, text, voice)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	wavData, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("pockettts: read response body: %w", err)
+	}
+
+	// ── Transcode WAV → MP3 via ffmpeg ────────────────────────────────────────
+	mp3Data, err := wavToMP3(ctx, wavData)
+	if err != nil {
+		return nil, fmt.Errorf("pockettts: transcode to mp3: %w", err)
+	}
+	return mp3Data, nil
+}
+
+// StreamAudioMP3 posts to POST /tts and returns an io.ReadCloser that delivers
+// MP3 bytes as pocket-tts generates WAV frames. ffmpeg runs as a subprocess
+// with stdin connected to the live WAV stream and stdout piped to the caller.
+// The caller must always close the returned ReadCloser.
+func (c *httpClient) StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error) {
+	if text == "" {
+		return nil, fmt.Errorf("pockettts: empty text")
+	}
+	if voice == "" {
+		voice = "alba"
+	}
+
+	resp, err := c.postTTS(ctx, text, voice)
+	if err != nil {
+		return nil, err
+	}
+
+	// Start ffmpeg: read WAV from stdin (the live HTTP body), write MP3 to stdout.
+	cmd := exec.CommandContext(ctx,
+		"ffmpeg",
+		"-hide_banner", "-loglevel", "error",
+		"-i", "pipe:0", // WAV from stdin
+		"-f", "mp3", // output format
+		"-q:a", "2", // VBR ~190 kbps
+		"pipe:1", // MP3 to stdout
+	)
+	cmd.Stdin = resp.Body
+
+	pr, pw := io.Pipe()
+	cmd.Stdout = pw
+
+	var stderrBuf bytes.Buffer
+	cmd.Stderr = &stderrBuf
+
+	if err := cmd.Start(); err != nil {
+		resp.Body.Close()
+		return nil, fmt.Errorf("pockettts: start ffmpeg: %w", err)
+	}
+
+	// Close the write end of the pipe when ffmpeg exits, propagating any error.
+	go func() {
+		waitErr := cmd.Wait()
+		resp.Body.Close()
+		if waitErr != nil {
+			pw.CloseWithError(fmt.Errorf("ffmpeg: %w (stderr: %s)", waitErr, stderrBuf.String()))
+		} else {
+			pw.Close()
+		}
+	}()
+
+	return pr, nil
+}
+
+// ListVoices returns the statically known predefined voice names.
+// pocket-tts has no REST endpoint for listing voices.
+func (c *httpClient) ListVoices(_ context.Context) ([]string, error) {
+	voices := make([]string, 0, len(PredefinedVoices))
+	for v := range PredefinedVoices {
+		voices = append(voices, v)
+	}
+	return voices, nil
+}
+
+// postTTS sends a multipart POST /tts request and returns the raw response.
+// The caller is responsible for closing resp.Body.
+func (c *httpClient) postTTS(ctx context.Context, text, voice string) (*http.Response, error) {
 	var body bytes.Buffer
 	mw := multipart.NewWriter(&body)
 
 	if err := mw.WriteField("text", text); err != nil {
 		return nil, fmt.Errorf("pockettts: write text field: %w", err)
 	}
-	// pocket-tts accepts a predefined voice name as voice_url.
 	if err := mw.WriteField("voice_url", voice); err != nil {
 		return nil, fmt.Errorf("pockettts: write voice_url field: %w", err)
 	}
@@ -105,34 +197,12 @@ func (c *httpClient) GenerateAudio(ctx context.Context, text, voice string) ([]b
 	if err != nil {
 		return nil, fmt.Errorf("pockettts: request: %w", err)
 	}
-	defer resp.Body.Close()
-
 	if resp.StatusCode != http.StatusOK {
 		_, _ = io.Copy(io.Discard, resp.Body)
+		resp.Body.Close()
 		return nil, fmt.Errorf("pockettts: server returned %d", resp.StatusCode)
 	}
-
-	wavData, err := io.ReadAll(resp.Body)
-	if err != nil {
-		return nil, fmt.Errorf("pockettts: read response body: %w", err)
-	}
-
-	// ── Transcode WAV → MP3 via ffmpeg ────────────────────────────────────────
-	mp3Data, err := wavToMP3(ctx, wavData)
-	if err != nil {
-		return nil, fmt.Errorf("pockettts: transcode to mp3: %w", err)
-	}
-	return mp3Data, nil
-}
-
-// ListVoices returns the statically known predefined voice names.
-// pocket-tts has no REST endpoint for listing voices.
-func (c *httpClient) ListVoices(_ context.Context) ([]string, error) {
-	voices := make([]string, 0, len(PredefinedVoices))
-	for v := range PredefinedVoices {
-		voices = append(voices, v)
-	}
-	return voices, nil
+	return resp, nil
 }
 
 // wavToMP3 converts raw WAV bytes to MP3 using ffmpeg.