Bug in check_jacobian when called with input_names and input_data

Summary

check_jacobian uses different inputs for analytical and approx jacobian if called with input_data and input_names

Gemseo version

6.0.0

Platform info

Windows

Steps to reproduce

from numpy import array

from gemseo import configure_logger, create_discipline
from gemseo.problems.mdo.sellar.variables import X_1, X_SHARED, Y_2

if __name__ == "__main__":
    configure_logger()
    sellar_1 = create_discipline("Sellar1")
    input_data = {
        X_1: array([3.0]),
        X_SHARED: array([3.0, 3.0]),
        Y_2: array([3.0]),
    }
    sellar_1.check_jacobian(
        input_data=input_data,
        input_names=[Y_2],
    )

What is the current bug behavior?

The check_jacobian fails.

What is the expected correct behavior?

It should pass.

Possible fixes

This might be because in check_jacobian, linearize is called with input_data, but the approx jacobian is called with the default values for the variables that are not in input_names. Thus, the analytical and approximated jacobians are evaluated at different inputs.