Those configurations aren't "impossible" due to fundamental well tested physics like general relativity. They're "impossible" from the perspective of our understanding of the formation of planetary systems, which is understandably less well developed given how much more difficult it is to study since as a science, exoplanet detection is just 30 years old, with the majority of detections being less than 15 years old.

In comparison, general relativity is one of the most well tested theories in physics, having been undergoing rigorous testing for over a century, and regardless, the point stands that the results were close to the model, thus providing evidence for the model's validity.

General Relativity has pretty much nothing to do with planet formation.

Newtonian gravity is good enough for that problem, and the problems have nothing to do with not understanding gravity well enough. The problems have to do with things like complex chemistry in stellar accretion disks, how grains in these disks stick to one another to eventually form rocks, and so on.

Whether the problems with the models are with general relativity, Newtonian gravity or other issues, the point is that the models have problems.

Which models are you talking about?

The fact that some models somewhere have problems has no bearing on whether the image reconstruction for Sgr A* is valid.

You might as well be saying that if fluid dynamics models have struggled to help predict or explain CPU performance increases, we should have the same level of expectation about how helpful a fluid dynamics based model of aerodynamics will be, i.e. complete nonsense

Planet formation is a subset of system formation which is a subset of galaxy formation and all are within the same observational astrophysics domain.

CPU performance and aerodynamics are two completely unrelated domains.