Fix a logic of Weight::applyGradient in nntrainer/tesnor/weight.cpp

This commit fixes a logic in `Weight::applyGradient(double lr, Tensor &updated_grad)` in `nntrainer/tensor/weight.cpp`
Previous version fallbacks to `applyGradient(lr)`, which does not use `updated_grad`.
This version now use `updated_grad` to apply gradients for FP32 path

Signed-off-by: PJH6029 <[email protected]>

Author: PJH6029

nntrainer/tensor/weight.cpp

@@ -137,7 +137,8 @@ void Weight::applyGradient(double lr, Tensor &updated_grad) {
     quantizeWeight();
     return;
   } else {
-    return applyGradient(lr);
+    /** FP32 (or matching dtype) path: apply the provided updated_grad directly */
+    var->add_i(updated_grad, -lr);
   }
 }