Support for module spec with no args

[paxcema]

Following recent changes to the analysis block parser, we should be able to specify things like encoders and mixers (ideally all possible modules, though) without the need for initial args. If they are missing, we simply revert to a default set of them.

In practice, this would enable the following:

json_ai = json_ai_from_problem(train_df, problem_definition=pdef)
json_ai.model['args']['submodels'] = [ {"module": "LightGBM"} ]

Instead of:

json_ai = json_ai_from_problem(train_df, problem_definition=pdef)
json_ai.model['args']['submodels'] = [
    {
        "module": "LightGBM",
        "args": {
            "fit_on_dev": "True",
            "stop_after": 120
        }
    }
]

[paxcema]

Update: this is now partially addressed, with support for arg-less mixers.