Add voice output (text-to-speech) to HumanCLI for Go2

Issue

Add voice output capability to HumanCLI so the agent can speak responses instead of just displaying text. Validate that it works on the Unitree Go2 platform.

Requirements

• Implement text-to-speech using Python audio libraries (e.g., pyttsx3, gTTS, espeak, or similar)
• Add directly to HumanCLI module (dimos/agents/cli/human.py)
• Support audio output on Go2 (speakers/audio device)
• Handle audio device selection/configuration
• Test on actual Go2 hardware

Implementation Considerations

• Use lightweight TTS that runs on-device or can call external API
• Low latency for real-time responses
• Voice should be clear and understandable in robot environment
• Toggle for enabling/disabling voice output
• Option to stream long responses instead of waiting for full synthesis

Acceptance Criteria

• Voice output works on Go2
• Agent responses are spoken aloud
• Audio quality is acceptable for human understanding
• User can toggle voice on/off
• Works alongside text output (not exclusive)

Related

• Pair with: Voice input (STT) issue (#1272)

Start by reading dimos/agents/cli/human.py to understand how HumanCLI produces responses and where voice toggling can fit alongside text output. Then investigate the available Go2 speakers/audio device and validate a lightweight Python TTS approach on actual Go2 hardware. Done means spoken responses, text output, audio configuration, and voice enable/disable behavior all work on Go2.