test(v7): deterministic product gate, preset-matrix fixes, reality evidence refresh

* gate:v7-product: the multi-axis smoke is now a deterministic engine-wiring
  check (damage for ATK, potencyRate, barrierRate, recoveryRate, controlChance
  for CONTROL_POWER vs TENACITY) instead of 8-battle win-rate samples, which
  were underpowered and flipped between runs; battle-smoke samples doubled to 16
  paired mirrored seeds without loosening any threshold. Now 24/24 green.
* audit-v7-presets: order each representative pair by TargetTheta (the
  representatives carry different levels, so rarity order was the wrong key) and
  separate the distinct-team 3v3 smoke from the side-neutrality probe, which now
  mirrors identical teams. Hierarchy uphold rate 0.887, side neutrality PASS.
* Refresh the Phase-6 reality evidence built by the takeover process
  (37,440-battle graph, Bradley-Terry fit, dispersion, residuals, ridge
  calibration, BPv4 reality report, preset matrix) and the regenerated manifest
  and component catalog. referenceGeneralPower stays at 300 so every committed
  artifact is reproducible by the committed code; the measured finding is
  recorded in docs/V7-DELIVERY-REPORT.md.

Author: NBL Agent

RELEASE-MANIFEST.json

@@ -1055,8 +1055,8 @@
     },
     {
       "path": "scripts/gate-v7-product.js",
-      "size": 8295,
-      "sha256": "68822caad0afbcfb5d155f7204bfbf5ef889b5e8e5c4efe94c88135cf9c7245c"
+      "size": 10246,
+      "sha256": "8b15fc26e18e618d8f0e9127787a3233f0bb36f2148f6492a3b7774fcbb961c6"
     },
     {
       "path": "scripts/gen-mc.js",