Investigate (and improve?) how `xr_vectorize` deals with time series data

[robbibt]

While testing #549, I found that xr_vectorize will happily accept a multi-temporal xarray dataset (e.g. multiple time steps), and return a single GeoDataFrame in return. The problem is I can't work out exactly how it is doing that: is it combining all layers along the time dimension, or just vectorising the first timestep?

We need to better understand how xr_vectorize deals with multi-temporal data, and potentially expand the function to do useful things like: take a multidimensional dataset, and vectorise each timestep individually (saving the results as rows labelled with time in the GeoDataFrame.