feedback6: merge R21P — every operator at full potential (潜能 6; GitHub #252, PR #255 by @SpiritedAwayCN, the owner's decision of 2026-10-07)

- tools/build-data.mjs applies every potentialRanks attribute step (COST, RESPAWN_TIME, ATK, MAX_HP, DEF,
  MAGIC_RESISTANCE, ATTACK_SPEED) to the chess, elites, 补位 and 自选 units and picks talent / trait / module-talent
  candidates up to requiredPotentialRank 5; summons take the owner's potential for their talents, not its stat steps.
  data/chess.json, backups.json, tokens.json and the en / ja / ko / zh-TW data translations rebuilt; op-wang.js reads
  her stock cap (8 at full potential). 刺玫 435 ATK / cost 15, 砾 16 s, 至简 70 s.
- No conflicts outside the golden files (regenerated). Golden vs HEAD: roster 49, bonds 46, fields 22, matches 18,
  diy 133 (standins and diy-012 / diy-135…138 unchanged: no potential ranks or never hit). Union check against the
  branch and the merge base: every scenario only the branch moved equals the branch, except roster-048 and
  solo-ABYSS-2: at full potential a 频次 器物 is reached there (roster-048: one more enemy dies instead of leaking, 86 → 87
  units), so R21H's hit count (data × hpMul ÷ supplyHpMul) applies where it did not before — with only that rule put back
  (archetypes.js times as at the merge base) both equal the branch exactly.
- Tests: broad node --test 5147 / 5147; lint 0 errors.

test/match/feedback5-we2-placement.test.js

@@ -65,7 +65,7 @@ test('#22 望\'s 棋子: a hand piece goes on a 高台 (ranged tile) as well as
   assert.equal(ps.gd.token(STONE).position, 'ALL');
   const wang = placeDiy(m, ps, T5A);
   const stack = ps.hand.find((p) => p && p.kind === 'token' && p.id === STONE && p.ownerUid === wang.uid);
-  assert.ok(stack && stack.count === 6, 'her 6 棋子 joined the hand');
+  assert.ok(stack && stack.count === 7, 'her 7 棋子 joined the hand (6 + 1 at full potential)');
   const high = tilesOf(ps, 'ranged');
   const ground = tilesOf(ps, 'melee');
   assert.ok(high.length && ground.length, 'the stage has both tile classes');