LearningRateControl, make sure lr is float

It might happen to be np.float64 or so...
It's not really such a big problem,
but it's a bit inconsistent then in the LR file.

returnn/torch/updater.py

@@ -92,7 +92,7 @@ def __init__(self, *, config, network, device, initial_learning_rate=1.0):
         :param float initial_learning_rate:
         """
         self.config = config
-        self.learning_rate = initial_learning_rate
+        self.learning_rate = float(initial_learning_rate)
         self._effective_learning_rate = self.learning_rate
         self.network = network
         self._device = device
@@ -150,7 +150,7 @@ def set_learning_rate(self, value):
 
         :param float value: New learning rate.
         """
-        self.learning_rate = value
+        self.learning_rate = float(value)
         self._update_effective_learning_rate()
 
     def get_effective_learning_rate(self) -> float:
@@ -162,9 +162,10 @@ def get_effective_learning_rate(self) -> float:
     def _update_effective_learning_rate(self):
         self._effective_learning_rate = self.learning_rate
         if self.learning_rate_function is not None:
-            self._effective_learning_rate = self.learning_rate_function(
+            lr = self.learning_rate_function(
                 global_train_step=self._current_train_step, epoch=self._current_epoch, learning_rate=self.learning_rate
             )
+            self._effective_learning_rate = float(lr)
         if self.optimizer:
             for param_group in self.optimizer.param_groups:
                 param_group["lr"] = self._effective_learning_rate