fix: F5-1 — initialize self._tokenizer before _ensure_tokenizer()

In __init__(), add self._tokenizer = None before calling self._ensure_tokenizer().
This fixes AttributeError: 'FusionModelInference' has no attribute '_tokenizer'.
8/8 tests passed.

inference/dashboard.py

@@ -77,6 +77,7 @@ class InferenceEngine:
         self.model.eval()
         self.inference_config = InferenceConfig()
         self.kv_cache = None
+        self._tokenizer = None
         self._ensure_tokenizer()
         
     def _load_model(self, model_path: str) -> tuple: