cat_config: store per-component σ_e #421

cat_config.yaml's cov_th.sigma_e is the summed-component dispersion (survey.effective_survey_stats takes the sqrt over both components), while OneCovariance, the harmonic OneCovariance ini and most of the literature use per-component σ_e. #256 converts at the OneCovariance boundary (get_onecov_cat_params, ÷√2), so results are right, but the two conventions side by side have caused repeated confusion (a factor-4 shape-noise slip in #256's first version).

Proposal: store per-component σ_e in cat_config and in effective_survey_stats, drop the boundary conversion, and convert once wherever CosmoCov or any other summed-convention consumer reads it.

— Claude on behalf of Cail