[Feature] Allow for actions and rewards to be in the reset tensordict (#458)

Author: vmoens

torchrl/envs/common.py

@@ -367,8 +367,8 @@ def reset(
             tensordict_reset = step_tensordict(
                 tensordict_reset,
                 exclude_done=False,
-                exclude_reward=True,
-                exclude_action=True,
+                exclude_reward=False,  # some policies may need reward and action at reset time
+                exclude_action=False,
             )
         if tensordict is not None:
             tensordict.update(tensordict_reset)