MSTL: handle non-seasonal data

MSTL currently bails out if passed non-seasonal data or periods. The [original paper](https://arxiv.org/abs/2107.13462) and R/statsforecast implementations, however, use `supersmoother` to smooth the trend before modelling that as usual.

We should do the same. We'll need to write a supersmoother implementation though...