Canonical survey area & effective number density helpers (dedupe sp_validation / shear_psf_leakage)

Effective number density and survey area are currently computed independently in (at least) two places:

• `shear_psf_leakage/rho_tau_cov.py`: `CovTauTh.get_area` (healpix-based) and `get_effective_number_density` — used internally for the tau theory covariance (CosmoStat/sp_validation#295 / shear_psf_leakage#35).
• `sp_validation`: `cosmo_val/catalog_characterization.py` (`calculate_n_eff_gal`, now per tomographic bin for the OneCovariance config, CosmoStat/sp_validation#298) and `survey.py:get_area`.

If the definitions drift (e.g. weighted vs raw n_eff, area estimators), the tau covariance and the OneCovariance C_ell covariance can silently disagree on shape noise and number density.

Proposal: one canonical implementation in cs_util that both packages call. Two composable options:

1. cs_util owns the calculation; sp_validation and shear_psf_leakage both import it.
2. Downstream code computes these once (in sp_validation, which owns the catalogue view) and passes them as explicit arguments to the covariance backends.

These are compatible — the calculation lives in cs_util, sp_validation calls it once and threads the values through. Open to opinions on the interface.

— Claude, on behalf of Cail

Start by comparing shear_psf_leakage/rho_tau_cov.py, especially CovTauTh.get_area and get_effective_number_density, with sp_validation/cosmo_val/catalog_characterization.py and survey.py:get_area. Read the related issues for the current per-bin and covariance requirements, then clarify the interface choice. Done means both downstream packages use consistent survey-area and effective-number-density values without duplicated definitions.