Added submenu for simpler loading of pre-trained models

[dribnet]

Added a small sub-routine that scans for previously
saved models. These can be loaded by simply selecting
the menu option instead of providing the full path.
The menu is presented in reverse chronological order
and the final option "Specify paths manually" preserves
the ability to provide a full path for the checkpoint
and replay buffer as was done previously.

This is probably best understood by looking at an example. Showing user input in bold, previously the UI flow for loading a model was:

0. Train
1. Load pretrained model
2. Diagnose model
3. Render some self play games
4. Play against MuZero
5. Test the game manually
6. Hyperparameter search
7. Exit
Enter a number to choose an action: 1
Enter a path to the model.checkpoint, or ENTER if none: results/cartpole/2020-12-28--21-53-13/model.checkpoint
Enter a path to the replay_buffer.pkl, or ENTER if none: results/cartpole/2020-12-28--21-53-13/replay_buffer.pkl

Using checkpoint from results/cartpole/2020-12-28--21-53-13/model.checkpoint
Initializing replay buffer with results/cartpole/2020-12-28--21-53-13/replay_buffer.pkl
Done

And after changes it becomes:

0. Train
1. Load pretrained model
2. Diagnose model
3. Render some self play games
4. Play against MuZero
5. Test the game manually
6. Hyperparameter search
7. Exit
Enter a number to choose an action: 1

0. results/cartpole/2021-01-03--01-53-29/
1. results/cartpole/2020-12-28--21-59-16/
2. results/cartpole/2020-12-28--21-58-40/
3. results/cartpole/2020-12-28--21-53-13/
4. Specify paths manually
Enter a number to choose a model to load: 0

Using checkpoint from results/cartpole/2021-01-03--01-53-29/model.checkpoint
Initializing replay buffer with results/cartpole/2021-01-03--01-53-29/replay_buffer.pkl
Done

Just an idea, no worries if you don't want to merge this into master as I can also appreciate a dead simple demo program.

[werner-duvaud]

LGTM, thanks !