gammapy: Investigate fail in FluxPointsEstimator

Gammapy version 1.0, 1.0.1

While running FluxPointsEstimator with multiple models present, and reoptimize=True, a confusing error has been sometimes reported

File ~/Gammapy-dev/gammapy/gammapy/datasets/evaluator.py:301, in MapEvaluator._compute_flux_spatial(self)
    299     value = (values.quantity * weights).sum(axis=(1, 2), keepdims=True)
    300 else:
--> 301     value = self._compute_flux_spatial_geom(self.geom)
    303 return value

File ~/Gammapy-dev/gammapy/gammapy/datasets/evaluator.py:309, in MapEvaluator._compute_flux_spatial_geom(self, geom)
    307 if not self.model.spatial_model.is_energy_dependent:
    308     geom = geom.to_image()
--> 309 value = self.model.spatial_model.integrate_geom(geom)
    311 if self.psf and self.model.apply_irf["psf"]:
    312     value = self.apply_psf(value)

File ~/Gammapy-dev/gammapy/gammapy/modeling/models/spatial.py:245, in SpatialModel.integrate_geom(self, geom, oversampling_factor)
    243     # Finally stack result
    244     result._unit = integrated.unit
--> 245     result.stack(integrated)
    246 else:
    247     values = self.evaluate_geom(wcs_geom)

File ~/Gammapy-dev/gammapy/gammapy/maps/wcs/ndmap.py:878, in WcsNDMap.stack(self, other, weights, nan_to_num)
    876     cutout_slices = Ellipsis, slices[0], slices[1]
    877 else:
--> 878     raise ValueError(
    879         "Can only stack equivalent maps or cutout of the same map."
    880     )
    882 data = other.quantity[cutout_slices].to_value(self.unit)
    883 if nan_to_num:

ValueError: Can only stack equivalent maps or cutout of the same map.

I tried to produce a minimal notebook here where the error shows (on cell 24) https://github.com/AtreyeeS/test_notebooks/blob/master/fpe_fail.ipynb

from what i understood, the wcs_geom here (https://github.com/gammapy/gammapy/blob/18c2597f23816f465189f6dde4e6e75fd9e50fe6/gammapy/modeling/models/spatial.py#L227) ends up with width : 0.0 deg x 0.0 deg when the spatial model position is very close to the lat/lon allowed range, probably leading to a situation with very little overlap on geom

The error is rather unstable, and sometimes vanishes on re-running the script.

It would be nice to handle this error better, but I dont know how.