fix: drop unused llama_cpp import in GGUF cross-encoder dependency check

Replaced the unused top-level import in `_check_llama_cpp` with `__import__("llama_cpp")`. This maintains the optional dependency check and custom error message while satisfying linting rules (F401).

Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>

Author: n24q02m

qwen3_embed/rerank/cross_encoder/gguf_cross_encoder.py

@@ -67,7 +67,7 @@
 def _check_llama_cpp() -> None:
     """Check that llama-cpp-python is installed."""
     try:
-        import llama_cpp  # type: ignore[unresolved-import] # noqa: F401
+        __import__("llama_cpp")
     except ImportError as e:
         msg = (
             "llama-cpp-python is required for GGUF models. "