TST: Improve power spectrum testing

[itrharrison]

The current test of classy power spectra fails, due to the interpolation creating >% level errors with respect to the previous function implementation, depending on the redshift.

We need to decide how to define a 'truth' power spectrum in order to create test data to validate which, if either of these is more correct.

This would probably require:

• [ ] creating a baseline P(k,z) where we had consistency at the 10^-4 level between camb and class run natively
• [ ] using this P(k,z) as the test data for our implementations which include interpolation

[itrharrison]

Two relevant resources: https://github.com/LSSTDESC/CCL/blob/master/benchmarks/test_camb_class_consistent.py https://github.com/xzackli/Cosmology-Notebooks/blob/master/CLASS%20vs.%20CAMB.ipynb