Change error to warning when particular solution to linear constraints does not satisfy them to the specified error threshold

[dpanici]

Right now we have a pretty tight error tolerance when we check that a set of linear constraints is satisfiable (

DESC/desc/objectives/utils.py

Line 159 in 50e805d

np.testing.assert_allclose(

) and we throw an error if this assert fails.

However there are cases where the error is on the order of say rtol=1e-13, which can happen just due to poor scaling when different constraints have highly varying orders of magnitude, or when the geometric scale of an equilibrium becomes very large. We probably want the solver to continue in these cases, so it might be better to use a warnif statement here instead of throwing an error, so that the user is aware then this is violated but then can use their judgement to see if it is because two fundamentally incompatible constraints were used (in which case error would be very large), or if it is just due to scaling issues between the variables (in which case the rtol would still be quite low)