Clearly document if/how irregular sampling is supported.

[johann-petrak]

I found this really hard to find out when one just tries to figure out if NeuralProphet should get considered to get tried out or not.

When I just ran exactly the same code as with Prophet with Neuralprophet, passing it irregular timestamps in the training series and the for forecasting, I got:

ValueError: Detected multiple frequencies in the timeseries please pre-process data.

So it looks like unlike Prophet, Neuralprohet cannot handle this?