Missing data in case of bilinear resampling of Lon/Lat data.

Steps to reproduce:

run Test_SensorMapGeometryTransformer.test_to_map_geometry_lonlat() with resampling algorithm 'blinear' -> pyresample raises a numpy warning due to NaN values which are NOT in the input coming from sensormapgeo:

/root/mambaforge/envs/ci_env/lib/python3.10/site-packages/pyresample/bilinear/_base.py:388
invalid value encountered in true_divide

The result looks like this:

Setting the radius_of_influence parameter to a very large value, reduces the missing pixels but does NOT remove them:

result = \
    NumpyBilinearResampler(source_geo_def, target_geo_def, **{'radius_of_influence': 30e300})\
    .resample(data, **{'fill_value': -9999, 'nprocs': 32})\
  .astype(data.dtype)

from geoarray import GeoArray
GeoArray(result).show()

Seems like this is a bug in pyresample.