Derive reporting xi means from an exactly nested fine grid

[cailmdaley]

Reporting ξ± means come from pair-weighted rebinning of one unpatched fine-grid measurement on exactly nested edges. The patched reporting measurement supplies covariance only and uses TreeCorr's default bin_slop; b_target governs the mean-producing pass.

• rebin_gg_means reuses pure-E/B's pair-weight operator (gg.weight) for ξ± and separations; it sums pair counts and weights. The route handles tomographic auto/cross pairs, and the reporting Snakemake job consumes the fine SACC product.
• Both shipped paper configs use k=34, 0.07939053012074941–301.3330747561176′, 1015 bins. The fine-grid means pass has bin_slop=1; configuration and consumption reject non-nested edges. COSEBIs and pure-E/B must be regenerated for the changed integration grid.
• Exact-counting weighted synthetic catalogues reproduce direct reporting TreeCorr means to 10⁻¹². Layout, cross-pair, producer and DAG tests pass.
• Full container suite at the stacked Derive reporting xi means from an exactly nested fine grid #400 head, slow tests included: 537 passed; three environment failures reproduce on T (seven missing configured paths and two NFS FITS-cleanup errors). Host workflow suite: 11 passed, 1 skipped.
• SP_v1.4.6.3: published mean, count and weight fields are bit-identical across seeded and develop-saved patch centres; the covariance differs by layout. Against unpatched reporting bin_slop=0.036: 0.095σ rms / 0.221σ max (angle_slop=0.01 reference: 0.053σ / 0.131σ). Against the old patched means: 0.299σ / 0.866σ seeded and 0.358σ / 1.119σ saved centres.
• Reusing nested_fine.npz, the default-tolerance covariance-only reporting pass took 3.95 min seeded / 3.76 min saved centres at 8 threads, excluding catalogue loading and patch-centre construction. The earlier fine measurement took 179.6 min once; it was not rerun for this timing.

Depends on #399; base is fix/layout-independent-means. The effect of changing bin_slop on jackknife covariance remains open in #401.

The decision record on assay/decision-record is not edited: this changes the real_space.integration_grid Values pins for cosmo_val.integration.{min_sep,max_sep,nbins} and bmodes/config/config.yaml#fiducial.{min_sep_int,max_sep_int,nbins_int}. bmodes.pure_eb_rebinning retains grid=integration and Σw_iw_j (gg.weight).

— Luna on behalf of Cail