Add Audio8 TTS engine integration

[diodiogod]

Summary

• add Audio8 TTS Preview 0.6B as a unified Text and SRT engine
• support reference-free generation, exact-transcript voice cloning, character voices, caching, inline tags, model download, and shared Transformers 4 isolation
• harden Audio8/Step Audio EditX unload-reload behavior and normalize post-edited audio before chunk assembly
• document Audio8 capabilities, license, model layout, and the isolated-worker import-path pitfall found during testing

Current status

Important

Audio8 is functional and produces valid audio in ComfyUI, but engine implementation testing is not finished. This PR is intentionally a draft until broader manual coverage is complete.

Validation completed

• live reference-free and exact-transcript cloned speech generation
• character switching and cache reuse
• Audio8 → Step Audio EditX → Audio8 lifecycle, including repeated unload/reload
• inline emotion/paralinguistic edits and mixed 2D/3D segment assembly
• focused runtime/audio-processing suite: 31 passed
• full unit suite: 143 passed, with 3 known unrelated failures

Remaining testing

• broader greedy/sampling parameter combinations
• SRT timing modes and longer subtitle sets
• long-form chunking and interruption/cancellation
• clean-install/model-download flow and additional hardware/runtime coverage
• Transformers 5 compatibility remains future work; the current known-working implementation uses the shared Transformers 4 profile

Scope note

Audio8 is not added to Voice Designer in this implementation. The official checkpoint does not expose an instruction-conditioned voice-design mode, and cloning instead requires reference audio paired with its exact transcript.