Default `fit_sum_model_variogram` can overfit short ranges

[rhugonnet]

See for example fit below, courtesy of Julian Llanto:

We should better adapt default arguments, maybe prevent a range shorter than X times the resolution? And prevent a fit within X times that of another range? (by some iterative process re-doing the fit several times and adjusting the bounds?) Not sure what the most robust method would be here :thinking:.

[erikmannerfelt]

For the moment, would it be possible to define threshold for when it becomes a problem and then trigger a warning if someone tries anyway?

[rhugonnet]

It doesn't seem to happen so often from what I hear... so maybe we can leave as is for now, until we add an enhancement.