Fix issue 49

[paolap]

To solve issue #49 caused by changes to the latest xarray:

• In land_check function when stacking dimensions land_check doesn't create anymore a multi-index ( this potentially should also speed up calculations), coordinates are added again before concatenating results.

In both threshold() and detect() was necessary to re-assign the coordinates, reset the "stacked" array using set_xindex to be able to unstack again, for example:

   dims = list(ts.cell.coords)
   mhw_results = [r[0].assign_coords({d: r[0][d][0].values for d in dims}) for r in results[0]]
    mhw = xr.concat(mhw_results, dim='cell')
    mhw = mhw.set_xindex(dims)
    mhw = mhw.unstack(dim='cell')

• As after groupby.aggregate operation coordinates are completely lost, they need to be re-added in the aggdf() function.
• Finally the check_coordinates() function in stats.py had to be adapted as now the stacked dimension is a dimension without coordinate

[paolap]

I also made some changes around data frames (in features.py) to avoid pandas warnings