[NEW FEATURE] Rho statistics: Flexible use of either training or test star sample

[martinkilbinger]

We should have the option to use either training or test star sample for rho statistics.
Only the second option is implemented by default.

Probably only have to change some config setting.

Closed by #636

[cailmdaley]

Heads up: this is likely obsolete within ShapePipe pending PR #715, which removes the rho-stats code path from the repo entirely (per the discussion on #657 — rho-stats now live in sp_validation / cosmo_val via shear_psf_leakage).

If the training-vs-test star-sample choice is still wanted, it would be a feature request against shear_psf_leakage.rho_tau_stat.RhoStat now, not shapepipe.

Flagging so you can decide whether to close or redirect once #715 lands. (Noticed while triaging after #715; drafted with Claude.)

[martinkilbinger]

Can now be closed.