problematic ets and stlm forecasts

[dashaub]

ets() ignores seasonality for long-seasonal periods, e.g. plot(forecast(ets(taylor), h = 500)) These forecasts are really bad since we're not really estimating an ets model.

stlm() will produce an error on a series that is not periodic, e.g. set.seed(123); stlm(rnorm(100)). We could accept this error and concede that the user should manually remove the s model, or we could drop the ets/stlm models in these cases (and throw a warning). In the case of ets, it is easy to check if frequency > 24. Similarly for stlm(), we can easily check if it is a ts object with at least two periods. Both approaches have advantages/disadvantages.

[dashaub]

Fixed in b9d9176
If either of these conditions occur, the ets/stlm model will be ignored, and the ensemble will run without these problematic models.