Add custom kq scaling from Gemma2Attention

convert-hf-to-gguf.py

@@ -2369,6 +2369,9 @@ class Gemma2Model(Model):
         self.gguf_writer.add_final_logit_softcapping(
             self.hparams["final_logit_softcapping"]
         )
+        self.gguf_writer.add_query_pre_attn_scalar(
+            self.hparams["query_pre_attn_scalar"]
+        )
 
     def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
         del bid  # unusem