Re-point paper/poster scaling-figure generators off the superseded cross-store join (#1901 finding)

Re-point the two scaling-figure generators that still render the superseded cross-store join, per #1901's supersession record eval_results/issue_1901/paper_densify/superseded_cross_store_join.json (committed on origin/issue-1901-mlpdense at 5a3df5a578, mirrored in the round's fold analysis).

Goal

scripts/issue1901_body_figures.py (fig_paper_c1_scaling, feeding figures/paper/c1_scaling_train_pool.png) and docs/posters/mats_2026/make_plot1_scaling.py still draw the scale7-store ≤25k scaling points (scaling_ladder_L19.json, mlp_scaling_L19.json) joined with the n1m ≥50k points (scaling_bigN_acc1_L19.json) as ONE ladder. #1901's mlp-scaling-densify round measured (2026-08-25, on-pod, epm:progress 2026-08-25T03:18:29Z + artifacts/mlpdense-smoke-r3-g1fail.log) that the two stores' pinned eval pools are DIFFERENT rows — 0/400 scale7 val rows content-match any of the 5,000 pass_b rows (median NN distance 6.63) — so any figure joining scale7 cells with n1m cells on one axis compares curves scored on different held-out rows.

Refuting evidence + replacement: the within-store dense ladder eval_results/issue_1901/paper_densify/mlp_scaling_dense_L19.json (all 8 fresh rungs 5k-500k drawn from the n1m store, one pinned val_400/test_1000 pool, whitened-cosine+CSLS primary retrieval; verdict Confirmed, D_gap = +0.0203). Reference rendering: scripts/issue1901_mlpdense_fold_figures.py (same branch) → figures/issue_1901/mlp_scaling_dense_L19.png.

Required changes:

1. fig_paper_c1_scaling: consume mlp_scaling_dense_L19.json for the linear + nonlinear curves (or render the old join ONLY with an explicit superseded/different-eval-pool label per the supersession record's disposition); keep the boundary-token control curves (the #1901 boundary round's contribution to the same figure) intact; re-render + commit figures/paper/c1_scaling_train_pool.{png,pdf,meta.json}.
2. docs/posters/mats_2026/make_plot1_scaling.py: same re-point for the poster plot; NOTE the poster is user-co-edited — regenerate the plot asset only, do not restructure the poster.
3. Preserve the boundary round's figure regression-gate convention where applicable (the prior gate pinned 304 committed points — the join replacement legitimately changes those points; supersede the gate's reference set deliberately, never silently).

Consumers list + disposition: see superseded_cross_store_join.json (consumers, disposition). Also check docs/posters/mats_2026/csls_rescore.py (validates vs scaling_ladder_L19.json) and docs/methodology/issue_1901.md per that record.