🩹 Fix missing matrix scaling in OptimizationMatrix.from_linked_data

Using dataset_scale (defined in scheme at experiment level for each dataset) to scale the optimization matrix when dealing with linked dataset.

This is on top of undoing the scaling of the raw data in the previous commit.

glotaran/optimization/matrix.py

@@ -196,7 +196,7 @@ def from_linked_data(
         return [
             cls.link(
                 [
-                    matrices[label].at_index(index)
+                    matrices[label].at_index(index).scale(linked_data.scales[label])
                     for label, index in zip(
                         linked_data.group_definitions[linked_data.group_labels[global_index]],
                         linked_data.data_indices[global_index],
@@ -268,12 +268,12 @@ def weight(self, weight: ArrayLike) -> OptimizationMatrix:
             self.array = self.array[np.newaxis, :, :] * weight.T[:, :, np.newaxis]
         return self
 
-    def scale(self, scale: float) -> OptimizationMatrix:
+    def scale(self, scale: float | Parameter) -> OptimizationMatrix:
         """Create a matrix container with a scaled matrix.
 
         Parameters
         ----------
-        scale : float
+        scale : float | Parameter
             The scale.
 
         Returns