make sure DefaultTrainerDict is pickle-able

[chriselion]

Proposed change(s)

In the current form, the DefaultTrainerDict can't be deserialized from pickle (and thus can't be passed to a worker process from the subprocess module, which I'm trying to do in #4780). It appears that the initializer needs to be able to take both forms of arguments; I'm not sure if there's a cleaner way to do this.

Useful links (Github issues, JIRA tickets, ML-Agents forum threads etc.)

Added in #4448

Types of change(s)

• Bug fix

Checklist

• Added tests that prove my fix is effective or that my feature works
• Updated the changelog (if applicable)

[chriselion]

Without the fix, the added test fails:

    def test_pickle():
        # Make sure RunOptions is pickle-able.
        run_options = RunOptions()
        p = pickle.dumps(run_options)
>       pickle.loads(p)

ml-agents/mlagents/trainers/tests/test_settings.py:526: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = DefaultTrainerDict(<class 'mlagents.trainers.settings.TrainerSettings'>, {})
args = (<class 'mlagents.trainers.settings.TrainerSettings'>,)

    def __init__(self, *args):
>       super().__init__(TrainerSettings, *args)
E       TypeError: 'type' object is not iterable

ml-agents/mlagents/trainers/settings.py:684: TypeError

[ervteng]

Maybe add a note about how it addresses an issue with pickle load for posterity

[chriselion]

Done

[chriselion]

only change between the last commit and the previous one was adding a comment, so going to merge without tests finishing.