Mlgame

**Changed**

* The recorded game progress for the inactivated ml client will be an empty list.

**Added**

* Add `"record_format_version"` field to the log file

**Changed**

* Change the data structure received from or sent to the game class: use a dict to store the information of each ml client. For example, the game has two players and defines the name of them as `"ml_1P"` and `"ml_2P"` in the `config.py`.
* The scene information returned from `get_player_scene_info()` will be:

{
"ml_1P": scene_info_for_ml_1P,
"ml_2P": scene_info_for_ml_2P
}

* The command sent to the `update()` will be:

{
"ml_1P": command_sent_from_ml_1P,
"ml_2P": command_sent_from_ml_2P
}

* And the command returned from `get_keyboard_command()` should also be:

{
"ml_1P": command_for_ml_1P,
"ml_2P": command_for_ml_2P
}

* Record the scene information and command individually for each ml client in the log file
* Update the version of built-in games
* arkanoid: 1.0 -> 1.1
* pingpong: 1.1 -> 1.2
* snake: 1.0 -> 1.1

**Fixed**

* The game could not be closed when running non-python client

**Changed**

* `-i/--input-script` supports specifying the file in the subdirectory
* For example, `-i foo/ml_play.py` means the file is at `games/<game_name>/ml/foo/ml_play.py`

**Removed**

* Remove `--input-module` flag

**Fixed**

* Handle the exception of `SystemExit`
* Handle the situation of reseting the ml executor before the game ends
* The game executor will ignore the ready command.

**Changed**

* Optimize the checking of the received object from either side

**Fixed**

* Use wrong value to chack frame delay

**Changed**

* Modify the game "snake" for the game development tutorial
* The `ml_play_template.py` of the game "snake" contains simple rule-based algorithm.

**Changed**

* ML process doesn't send the command only when the returned value of `MLPlay.update()` is `None`.