Challenges v2 quality follow-ups (final review residuals)

[Daren9m]

Non-blocking residuals from PR 2's final whole-branch review (all draw-neutral text/pool work):

• Investigation: Hard × set-piece webs still hit duplicate clue-card pairs ~55/100 (witness pool of 5 wraps under 12 clues) — vary text on wrapped draws; also small pools for the investigation complication and red-herring templates; observation-oblique on the Who node can gently mislead (document as oblique-by-design or special-case).
• Trap: grim lasting-cost rider is a fixed string duplicating GRIM_RIDERS[1] — draw from a pool per spec §6.6; whimsical escalation can double 'each round' when the frame's escalation starts with it; TrapFrame could carry a save ability instead of hardcoded DEX.
• Chase: themed waypoint sub-pools (current pool is global and urban-leaning; spec §8.5 says themed).
• Skill challenge: per-skill narration hooks are generic; social Persuasion static dc deserves a comment (= Indifferent-stage DC).
• Chase trick round: triple em-dash stacking softening.
• Orchestrators: id omits theme/difficulty → same-seed same-type history dedupe collisions (both pages; see also Puzzles v2 code-health batch (final review follow-ups) #58's PR-1 note); consider a full-JSON content-hash golden pin so framework-internal draw reorders are caught (current pins lock orchestrator order only).
  Context: docs/superpowers/plans/2026-07-16-noncombat-engine-v2-pr2-challenges.md; ledger .superpowers/sdd (local).