voice-codelisted

# Voice Code Run a deliberate voice loop through the `mcp__voicebridge__voice_status`, `mcp__voicebridge__voice_models`, `mcp__voicebridge__voice_configure`, `mcp__voicebridge__voice_start`, `mcp__voicebridge__voice_speak`, `mcp__voicebridge__voice_listen`, and `mcp__voicebridge__voice_stop` tools. VoiceBridge only performs speech input and output. Codex controls the conversation and authors every word sent to speech. Maintain two outputs for completed work: - Keep useful technical detail in the normal written Codex response. - Compose a separate short, conversational summary for `voice_speak`. Never read code, file paths, bullet lists, or the full written response aloud. 1. Call `voice_status`. If `first_run` is true, call `voice_models` before any audio tool or model download. Present separate TTS and STT single-choice selectors using Codex's structured question UI when available, with a numbered conversational fallback. The structured UI starts on the first listed option, so list the returned default first in each selector (Pocket TTS and Parakeet 110M STT) with "(Recommended)" appended to its label, then the remaining options lightest to heaviest. Show each option's tier and download size. If the user cancels, end without calling `voice_start`. Otherwise call `voice_configure` with both selected IDs. Existing users with `first_run: false` skip this onboarding choice. 2. Call `voice_start` and wait for the audio preflight and local speech m