Add open_questions_gate scenario exercising mapping-form scripted gates

[PolyphonyRequiem]

What

Adds tests/harness/scenarios/open_questions_gate_answer_loop_then_abort/scenario.yaml — the first scenario to exercise the AB#3212 scripted-gate infrastructure's mapping form (gate value + additional_input payload).

Why

AB#3239 PR #482 shipped the first scripted-gate scenario (plan_level_closed_unmerged_acknowledges_via_gate), which used the string shorthand form. AB#3212 PR #483 added the mapping form so gates declaring prompt_for: X can carry an additional_input payload. Nothing exercised that path until now. open_questions_gate.answer declares prompt_for: answers — the canonical proving ground.

Path covered (plan-level.yaml)

depth_guard -> guidance_loader -> root_resolver -> type_loader
  -> ancestor_chain -> state_detector (state=not_started)
  -> architect (call #1: 1 high-severity open_question)
  -> research_loop_counter -> open_questions_policy
  -> open_questions_counter (iter 0 < cap; question gates)
  -> open_questions_gate (scripted=answer + additional_input.answers)
  -> architect (call #2: open_questions cleared by operator answer)
  -> research_loop_counter -> open_questions_policy
  -> open_questions_counter (no gating questions remain)
  -> ensure_plan_branch (returns error envelope)
  -> ensure_plan_branch_error_gate (scripted=abort, string shorthand)
  -> terminal_abort_run -> $end

The architect answer-loop round-trip (gate answer route -> back to architect -> open_questions cleared) is the headline behaviour; the ensure_plan_branch_error_gate abort is just the clean terminal so we don't have to drive the rest of the plan-PR lifecycle.

Why this scenario is worth its weight

• Only mapping-form coverage we have. Without it, scripted-gate additional_input is shipped but untested.
• Round-trip proof. The fake provider returns scripted output feat: Phase Detection, Routing Engine, and Command Implementations (PG-3) AB#2593 AB#2594 #2 only because the gate's answer route navigates back to architect — proving the gate routing actually fired.
• Two scripted gates in one scenario. Mapping form (open_questions_gate) AND string form (ensure_plan_branch_error_gate). With AB#3212 strict_gates=true, a typo in either name fails the scenario loudly.

Results

• Local single-scenario run: passed in 10.5 s.
• Local full Pester suite: 13/13 passing in 231.7 s.
• Zero direct git, zero LLM calls, all polyphony verbs faked through the shim.

Notes

• ensure_plan_branch returns error to short-circuit the downstream chain. In real runs, ensure_plan_branch_error_gate would more typically be hit on transient git failure — the abort path was already there; we just walked it.
• AB#3210 (pre_coder_branch_mismatch_gate) shipped in PR test(harness): pre_coder_branch_mismatch_gate abort scenario (AB#3210) #484. AB#3211 (squash_coverage_mismatch_gate) remains blocked on a per-scenario tmp git init fixture — separate ADO item to follow.

Closes: AB#3239 follow-up coverage of open_questions_gate mapping form.
Refs: AB#3212 (scripted-gate infrastructure), AB#3239 (initial closed_unmerged scenario).