Support dynamic inputs and outputs

[pvk-developer]

Problem Description

Currently MLPipeline can't reach the primitives inputs or outputs dynamically, which means that it will return the current value that is hold within the json file or the dictionary.

If we can specify a function that returns the expected input or output we will be able to dynamically read those even if they are not specified within the json file (the function should be).

Expected Behavior

1. The MLBlock,_extract_params should consider that self.fit_args and self.produce_args could be str pointing to a function. This would return empty dictionaries for the fit_params and produce_params.
2. When calling MLPipeline._get_block_variables if we recieve a str when we get the attributes for variables_attr we would call the block.instance with the returned value as a function using getattr:

if isinstance(variables, str):
    variables = getattr(block.instance, variables)()

Example of primitive with dynamic input and output

{
    "name": "mlblocks.MLPipeline",
    "primitive": "mlblocks.MLPipeline",
    "fit": {
        "method": "fit",
        "args": "get_fit_args"
    },
    "produce": {
        "method": "predict",
        "args": "get_predict_args",
        "output": "get_outputs"
    },
    "hyperparameters": {
        "fixed": {
            "pipeline": {
                "type": "str",
                "default": null
            },
            "primitives": {
                "type": "list",
                "default": []
            },
            "init_params": {
                "type": "dict",
                "default": {}
            },
            "input_names": {
                "type": "dict",
                "default": {}
            },
            "output_names": {
                "type": "dict",
                "default": {}
            }
        }
    }
}