Handling NaN's

[fealho]

DeepEcho 0.2.0 cannot handle NaN's. This means that if SDV passes a NaN value to DeepEcho, the current code crashes.

The snippet below reproduces this issue:

sequences = [
    {
        'context': [],
        'data': [
            [0.0, np.nan, 1.0],
            [0.5, 0.4, 1.0],
        ]
    }
]
context_types = []
data_types = ['continuous', 'continuous']

model = PARModel()
model.fit_sequences(sequences, context_types, data_types)
model.sample_sequence([])