Fix more indexing bugs

ml-agents/mlagents/trainers/optimizer/torch_optimizer.py

         ):
             seq_obs = []
             for _obs in tensor_obs:
-                start = seq_num * self.policy.sequence_length - leftover
-                end = (seq_num + 1) * self.policy.sequence_length - leftover
+                start = seq_num * self.policy.sequence_length - (
+                    self.policy.sequence_length - leftover
+                )
+                end = (seq_num + 1) * self.policy.sequence_length - (
+                    self.policy.sequence_length - leftover
+                )
-                assert _obs[start:end].shape[0] == self.policy.sequence_length
             values, _mem = self.critic.critic_pass(
                 seq_obs, _mem, sequence_length=self.policy.sequence_length
             )