Rewards-to-Go issue

[Raymondliz]

When rewards-to-go is implemented in the on policy ALGOBuffer.finish_path(), looks all necessary info has been recorded in the rew_buf. The last_val for rews seems just a placeholder because the action has been sent out to agent and the reward has been seen and recorded.
The last_val for value function totally makes sense for bootstraping but reward?
My thinking is the rewards-to-go might be this

self.ret_buf[path_slice] = core.discount_cumsum(rews[:-1], self.gamma)

instead of

self.ret_buf[path_slice] = core.discount_cumsum(rews, self.gamma)[:-1]