feat: persist voice samples to disk, generate VTT, cleanup after analysis

internal/domain/entity/voice_sample.go

@@ -0,0 +1,17 @@
+package entity
+
+import "time"
+
+// VoiceSample tracks all files produced during voice message processing.
+// Files are kept on disk until Cleanup() is called after successful analysis.
+type VoiceSample struct {
+	ID         string
+	UserID     int64
+	ChatID     int64
+	FileIDTg   string // original Telegram file ID
+	OGGPath    string // downloaded raw audio
+	WAVPath    string // ffmpeg-converted audio
+	VTTPath    string // WebVTT subtitle file generated by Whisper
+	Transcript string
+	CreatedAt  time.Time
+}