[RLlib] Fix use_lstm flag for ModelV2 (w/o ModelV1 wrapping) and add it for PyTorch. (#8734)

rllib/tuned_examples/ppo/repeatafterme-ppo-lstm.yaml

@@ -0,0 +1,24 @@
+repeat-after-me-ppo-w-lstm:
+    env: "ray.rllib.examples.env.repeat_after_me_env.RepeatAfterMeEnv"
+    run: PPO
+    stop:
+        episode_reward_mean: 50
+        timesteps_total: 100000
+    config:
+        # Works for both torch and tf.
+        framework: tf
+        env_config:
+          config:
+            repeat_delay: 2
+        gamma: 0.9
+        lr: 0.0003
+        num_workers: 0
+        num_envs_per_worker: 20
+        num_sgd_iter: 5
+        vf_share_layers: true
+        entropy_coeff: 0.00001
+        model:
+            use_lstm: true
+            lstm_cell_size: 64
+            max_seq_len: 20
+            fcnet_hiddens: [64]