release fixed scipy version by solving interpolation issue introduced since scipy 1.10.0

[veenstrajelmer]

There are xarray interpolation issues with scipy>=1.10.0:

• fix Interpolation difference between float and da with dims #287 and Recently introduced different behaviour of da.interp() when using floats vs DataArrays with new dim pydata/xarray#7701
• actual cause seems scipy version independent: BUG: RegularGridInterpolator on dataset with nans returns unexpected nans scipy/scipy#19318
• workaround possible or can we wait on interpn fix? >> seems to be working as expected, regulargridinterpolater works better than in 1.9.3. Maybe consider to accept the new behaviour? (first check initial cmems case in dfmt issue)

Time is running out since scipy<1.10.0 has memory leak

Also, Python 3.11 environment now gives a deprecationwarning which disappears when using scipy 1.10.0 and later versions:

C:\Users\veenstra\Anaconda3\envs\dfm_tools_py311\Lib\site-packages\scipy\sparse\_sputils.py:44: DeprecationWarning: np.find_common_type is deprecated.  Please use `np.result_type` or `np.promote_types`.
See https://numpy.org/devdocs/release/1.25.0-notes.html and the docs for more information.  (Deprecated NumPy 1.25)
  upcast = np.find_common_type(args, [])

MWE (dfmt testcase test_xugrid_opendataset_ugridplot_contourf_scipy_numpy_deprecation() implemented in #560):

import dfm_tools as dfmt
import xugrid as xu

file_nc = dfmt.data.fm_curvedbend_map(return_filepath=True)

uds = xu.open_dataset(file_nc)
uds['mesh2d_flowelem_bl'].ugrid.plot.contourf()

Also rerun postprocessing notebook

The unexpected nans in the interpolation to a new xarray dimension, occurs due to new behaviour of scipy's RegularGridInterpolator. This is not unexpected or undesired, but just inconvenient in this case. It seems to have to do with the ordering of the arrays, but flipping the input array does not solve it since there is some sorting going on somewhere in the code. interp1d was tested by looping over all points and concatenating the arrays, and also by interp1d on a grid of lat*lon. Both were significantly slower than interpn/RegularGridInterpolator. @Huite suggested to stick to the linear interpn, but follow-up with a nearest interpn ([NearestNDInterpolator](https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.NearestNDInterpolator.html)) and combine with combine_first. This does not seem to affect performance too much and it does solve the problem of too much nans.