How to Design a Fully Streaming Voice Agent with End-to-End Latency Budgets, Incremental ASR, LLM Streaming, and Real-Time TTS