[Enhancement] Add streaming TTS WebSocket client

[deepgram-robot]

Summary

Add a TTS WebSocket streaming client to the Go SDK, enabling real-time text-to-speech with chunked audio delivery — matching the TTS WebSocket support already available in the Python and JS SDKs.

Problem it solves

Go developers building voice agents, IVR systems, and real-time audio applications need streaming TTS for low-latency speech synthesis. The Go SDK currently supports TTS only via the REST endpoint (full audio response), which adds latency unacceptable for conversational use cases. A WebSocket TTS client would enable Go developers to receive audio chunks as they're generated, matching the experience in other Deepgram SDKs.

Proposed API

client := speak.NewWebSocketClient(apiKey)

err := client.Connect(speak.WebSocketOptions{
    Model:    "aura-asteria-en",
    Encoding: "linear16",
})

client.OnAudioData(func(data []byte) {
    // Play or relay audio chunk
})

client.SendText("Hello from Deepgram!")
client.Flush()
client.Close()

Acceptance criteria

• WebSocket client for TTS streaming endpoint
• Callback-based audio chunk delivery (matching Listen WS client pattern)
• Supports flush and close operations
• Supports all Aura voice models
• Documented with usage example
• Compatible with existing SDK architecture and error handling patterns

---

Raised by the DX intelligence system.