Add support for time series prediction

[iirekm]

Unfortunately, mljar seems to be inadequate for time series prediction. Missing features:

• optionally: some time-series specific feature extraction (e.g. rolling means, convolutional neural net, etc)
• optionally: quantile loss, ideally if multiple quantiles could be predicted (like Amazon Forecast or AutoGluon does)
• at support for TimeSeriesSplit - I've noticed that mljar has validation_type="custom", cv= but it seems to be unimplemented, although cv is provided, I get You need to specify cv as list or iterable