IGV normalization explanation: causal report + LegNet stub parity

Final pre-ship verification: 17 shipped HTML reports audited for
(a) IGV embedded, (b) signal tracks floor-rescaled to [0, 3.0],
(c) user-facing "noise floor (p95) / peak threshold (p99)"
explanation next to the IGV block, (d) track labels carry
assay:cell_type provenance (oracle · assay:cell_type for multi-oracle).

Two reports failed before this fix:

1. causal_prioritization/SORT1_locus causal report — 65/67 tracks
   were already floor-rescaled to ymax=3.0 (the 2 outliers are
   "Sentinel" + "Composite Score" summary tracks, correctly [0,1]),
   but the scale explanation block from variant_report.py wasn't
   replicated in causal.py. Added the same <p> note directly after
   the existing IGV intro paragraph (chorus/analysis/causal.py:1190).

2. validation/SORT1_rs12740374_multioracle/rs12740374_SORT1_legnet_report.html
   — stale from before the multi-oracle regen flow started
   persisting _normalizer through to_dict/from_dict.
   Re-ran scripts/regenerate_multioracle.py --oracle legnet (and
   --consolidate to keep the unified report in sync). LegNet stub
   now shows the floor/p99 block and uses the rescaled axis.

Result: 17/17 HTMLs pass the 4-part audit. Track provenance verified:
every single-oracle track label is {assay}:{cell_type}; every
multi-oracle track label is {oracle} · {assay}:{cell_type}. Every
IGV y-axis is either 3.0 (signal tracks, floor-rescaled) or 1.0
(summary/composite tracks, [0,1] by construction). No raw-autoscale
leaks.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: lucapinello

chorus/analysis/causal.py

@@ -1187,6 +1187,14 @@ def _build_causal_html(result: CausalResult) -> str:
                 'Signal tracks show ref (grey) vs alt (colored) for the top candidate. '
                 'Red stripe marks the sentinel. Zoom and pan to explore.</p>'
             )
+            p.append(
+                '<p style="font-size:.85rem;color:#6b7280;margin-top:-.5rem">'
+                "Signal rescaled using each track's genome-wide noise "
+                "floor (p95) and peak threshold (p99): "
+                '<b>0</b> = noise floor, <b>1.0</b> = top 1% of bins '
+                'genome-wide. Peak shape preserved; tracks comparable '
+                'across cell types.</p>'
+            )
             p.append(igv_html)
     except Exception as exc:
         logger.debug("Could not render IGV browser: %s", exc)