fix(backend): keep the trimmed clip when a runaway retry cannot split further; track the S3 tokenizer repo in model status and delete

Review follow-ups: with retries_runaway on for an engine that also has a
trim step, a <=100-char chunk flagged as runaway raised instead of
falling back to the trimmed output that already cuts the silence+noise
tail; generate_chunked now returns the trimmed chunk in that terminal
case. ModelConfig gains aux_hf_repo_ids so /models status only reports
the Chatterbox MLX model as downloaded once mlx-community/S3TokenizerV2
is present too (matching the backend's own cache check) and
DELETE /models/{name} removes that repo as well.

backend/backends/chatterbox_mlx_backend.py

@@ -20,6 +20,7 @@ from typing import ClassVar
 
 import numpy as np
 
+from . import CHATTERBOX_MLX_S3_TOKENIZER_REPO
 from .base import (
     combine_voice_prompts as _combine_voice_prompts,
     is_model_cached,
@@ -32,7 +33,7 @@ logger = logging.getLogger(__name__)
 CHATTERBOX_MLX_HF_REPO = "mlx-community/chatterbox-multilingual-v3"
 # mlx-audio's Model.from_pretrained fetches the S3 speech tokenizer from this
 # second repo (~470 MB), so the engine is only "downloaded" once both are cached.
-S3_TOKENIZER_HF_REPO = "mlx-community/S3TokenizerV2"
+S3_TOKENIZER_HF_REPO = CHATTERBOX_MLX_S3_TOKENIZER_REPO
 
 # Files that must be present for the MLX multilingual model
 _MLX_WEIGHT_FILES = ["model.safetensors", "config.json", "tokenizer.json"]