feat(multimodal): 视觉分层路由（多模态检测→主/视觉/OCR 工具）+ 语音 STT→TTS 降级链（异步）

internal/runtime/multimodal/voice/voice.go

@@ -0,0 +1,35 @@
+// Package voice implements asynchronous speech-to-text and text-to-speech for
+// the Cyrene runtime, following DevDocs/design/06-多模态.md §1.
+//
+// Voice input (STT) is transcribed to text and fed into the same agent pipeline
+// as a chat message; the reply is synthesized (TTS) back to audio. Both
+// directions degrade through a provider chain — cloud → local → edge-tts/espeak
+// → silent — so an unavailable provider falls through to the next rung instead
+// of failing the whole turn. The stub default implementations are HTTP-based and
+// testable with a stdlib httptest server; real cloud/local providers remain
+// adaptation points for later milestones.
+package voice
+
+import "context"
+
+// Audio is a piece of audio data plus its encoding.
+type Audio struct {
+	// Format is the encoding, e.g. "wav", "mp3", "ogg".
+	Format string
+	// Data is the raw audio bytes.
+	Data []byte
+}
+
+// STT converts audio to text.
+type STT interface {
+	// STT transcribes audio and returns the spoken text. ctx controls
+	// cancellation and must be honored.
+	STT(ctx context.Context, audio Audio) (string, error)
+}
+
+// TTS synthesizes speech audio from text.
+type TTS interface {
+	// TTS synthesizes audio from text. ctx controls cancellation and must be
+	// honored.
+	TTS(ctx context.Context, text string) (Audio, error)
+}