Update for libtorch 2.0

PyTorch 2.0 deprecated passing a int-valued mask to masked_fill.
This change updates one useage in dqn.cc to use bools.

open_spiel/algorithms/dqn_torch/dqn.cc

@@ -216,7 +216,7 @@ void DQN::Learn() {
     legal_actions_mask.push_back(
         torch::from_blob(t.legal_actions_mask.data(),
                          {1, t.legal_actions_mask.size()},
-                         torch::TensorOptions().dtype(torch::kInt32))
+                         torch::TensorOptions().dtype(torch::kBool))
             .clone());
     actions.push_back(t.action);
     rewards.push_back(t.reward);
@@ -231,7 +231,7 @@ void DQN::Learn() {
       next_info_states_tensor).detach();
 
   torch::Tensor illegal_action_masks_tensor =
-      1.0 - torch::stack(legal_actions_mask, 0);
+      torch::stack(legal_actions_mask, 0).bitwise_not();
   torch::Tensor legal_q_values =
       torch::masked_fill(target_q_values, illegal_action_masks_tensor,
                          kIllegalActionLogitsPenalty);