CFEL methods are not adequately tested

[hverdonk]

Per discussions in #11, we want more comprehensive cfel tests that cover new features introduced in #9 and #11. Ideally, the tests would cover the following new behavior for the CFEL method:

• process_results() now returns diff_sites as the number of rows with Q ≤ 0.20.
• process_comparison_data() returns, per group: group_N, group_T, group_dN/dS, group_aa_conserved.
• process_comparison_site_data() returns, per site and group: cfel_marker, cfel_beta, composition, substitutions, majority_residue.
• Per-group beta values are formatted as: 0.000 if exactly 0, scientific with 3 decimals if |beta| < 1e-3 or ≥ 1e3, fixed 4 decimals otherwise
• cfel_marker now reports Q-value formatted to 3 decimals when Q ≤ 0.20, otherwise -; malformed values yield NA.