Description

Voice input previously failed silently on models that don't support audio (e.g., Claude, GPT-3.5). This adds intelligent model capability detection so the mic button is only enabled for supported models (GPT-4o, Gemini), with clear user feedback when voice isn't available.

PRs

• #621: add model-aware voice input with audio support detection

Impact

Frontend (adk-web), Backend (OpenAI LLM audio handling)