Created by: skylian

Two changes need to be done: 1. In Experience, we've decided to let done correspond to the next state instead of the current state, as discussed in the design doc. 2. Pre-specify the shape of each field in Experience and pre-allocate space for them in ReplayBuffer for efficiency. After discussion with @yu239,

1. We've decided to keep the original choice: using Experience.game_status for the game status of the current state.
2. We've decided not to do pre-allocation for now