DHLearn: Scatterer: Cast first to PyObject before casting to provided type

[nbauernfeind]

JJ's DHLearn example doesn't work because this scatter method returns Object instead of PyObject. Thus the method that is picked is a Object -> (double) cast. Which means that the PyObject fails when attempting to be converted to Double.

If Scatterer is only to be used from python, then we could forcefully cast to PyObject before the type.

If Scatterer is to be used also from Groovy, then we might need to do some custom casting in scatter iff the result value is an instance of PyObject.

The important bits of JJ's example is this:

# A function to extract a result
def get_results(data, idx):
    return data[idx]

result = learn.learn(
    table = source,
    model_func = sum_columns,
    inputs = [learn.Input(["X", "Y"], table_to_array)],
    outputs = [learn.Output("Z", get_results, "double")],
    batch_size = 1000

Where that get_results method is treated as Function<Object[], Object> instead of Function<Object[], PyObject>.