feat: audio & 3D generation subcommands (generate audio / generate 3d, models --type audio|3d)

[binyangzhu000-sudo]

Gap

A capability audit against the live platform catalog (~350 live models, 13 categories) found the CLI's command surface covers only three channels — chat, generate image, generate video — which leaves two whole output modalities unreachable from the CLI:

Platform capability	Models	CLI today
Text-to-speech (TTS)	Seed Audio 1.0, xAI TTS, ElevenLabs	❌ no subcommand
Music generation	Suno Chirp family, MiniMax Music	❌ no subcommand
Speech-to-text (ASR)	Seed ASR 2.0, xAI STT	⚠️ only indirectly via chat --audio with a multimodal LLM — not the dedicated ASR models
3D (image/text-to-3D)	Seed3D 2.0, Hunyuan 3D	⚠️ works via generate image (3D models are Image-type) but is undiscoverable

Compounding this, atlas models list --type only accepts chat|image|video, so Audio-type models cannot even be listed from the CLI.

Proposal

1. atlas generate audio MODEL -p "..." — TTS + music; same submit/poll UX as image/video; --audio @file input passthrough for ASR models (output = transcript text).
2. atlas models list --type audio (and ideally 3d as a filter alias for the 3D categories within Image-type).
3. atlas generate cost audio for pricing parity.

All of these map to the existing POST /api/v1/model/generateAudio endpoint — the MCP server (v1.5.0) already drives TTS, music, and ASR through it, so this is CLI surface work only, no new backend.

Context

The repo README previously advertised "Audio — Grok TTS" which the binary cannot actually call; that's been corrected to describe audio-as-input honestly. This issue tracks closing the real gap. (Go source is maintained privately, hence an issue rather than a PR.)

🤖 Generated with Claude Code