remove some degenerate constraints from test covariance so that we don't try to invert an almost singular matrix; add separate test for nu constraint

The test of the overall AIMS run was using multiple frequency constraints that were degenerate. i.e. they contained duplicate information. This meant the covariance matrix was almost singular (the condition numbers were > 10¹⁷). So far, the test worked because finite-precision arithmetic still allowed np.linalg.inv to successfully invert the matrix but I got local failures presumably because of slightly different floating point calculations. I've removed the degenerate constraints in src/AIMS_configure.py.

To make sure that calculating the covariance matrix with the individual frequencies still works, I also added a small extra snippet to test those in src/tests/test_aims.py.