Refactor tests as suggested

[lauraporta]

          I'll approve for now in the interest of moving forwards, and as discussed, it might be worth opening an issue to restructure the tests. The proposed structure of `calculate_stats_outputs` that we came up with is that we'd rename it to `regenerate_regression_data` (or similar) and structure it something like:

def generate_regression_data(seeds):
    input_data = read_hdf5() # same function(s) that are used in the source code, not in `fixture_helpers.py` to read from MatLab/HDF5
    f = FrequencyResponsiveness(input_data)
    results = do_analysis(f)
    save_results(results) # so you can compare in your regression tests

This would include using an existing Matlab output file or generating it with h5py.

This has the added advantage of not relying on the same functionality as the fixtures, and therefore would reduce complexity in conftest.py a lot?

Originally posted by @alessandrofelder in #57 (comment)