fix(offline-guard): make every cache check cover the files the forced-offline load reads

The load body now runs with HF_HUB_OFFLINE forced when _is_model_cached()
reports True, so a snapshot that holds the weights but not the small files
the loader also opens would fail hard instead of fetching them. Verified
against chatterbox-tts 0.1.7 (mtl_tts.py allow_patterns, tts_turbo.py
from_local) and the TADA loader's unsloth/Llama-3.2-1B tokenizer download;
list those files in the required_files checks. Also note in the changelog
why the 0.4.5 removal of this guard no longer applies.

backend/backends/chatterbox_turbo_backend.py

@@ -29,11 +29,18 @@ logger = logging.getLogger(__name__)
 
 CHATTERBOX_TURBO_HF_REPO = "ResembleAI/chatterbox-turbo"
 
-# Files that must be present for the turbo model
+# Every file ChatterboxTurboTTS.from_local() reads (chatterbox-tts 0.1.7): the
+# three weight files, the GPT-2 tokenizer files AutoTokenizer needs, and the
+# built-in voice. The load runs with HF offline mode forced when this reports
+# cached, so a partial snapshot must not count as cached.
 _TURBO_WEIGHT_FILES = [
     "t3_turbo_v1.safetensors",
     "s3gen_meanflow.safetensors",
     "ve.safetensors",
+    "tokenizer_config.json",
+    "vocab.json",
+    "merges.txt",
+    "conds.pt",
 ]