nx_photogram_front_diag.nx

nx_photogram_front_diag.nx -- measurement probe, not a gate. T5 failed with the noise control scoring 18/20, which says the epipolar residual is not separating true correspondences from garbage. Before touching a tolerance constant I want the actual numbers: what does e3_residual return for a TRUE correspondence, for a planted outlier, and for pure noise -- both under an E built from the known pose (ep_build_E) and under an E estimated from the data (e3_estimate_norm). Tuning a threshold without looking at the distribution is how a gate gets faked green. license_tier: ORIGINAL expect_exit: 0