feat: mode-specific scene input rows for teaching + devotional (#1734)

[CraigBuckmaster]

Closes #1734. Phase 2.5 of epic #1725.

Why

Teaching and Devotional are the two modes whose output depends on the user's situation, not just the chapter. This card extends the Scene step to actually capture that situation (audience/setting for teaching; arrival for devotional). Quick and Deep stay information-only.

What

app/src/services/guidedStudy/types.ts — GuidedSceneRow becomes a discriminated union:

type SceneInputKey = 'audience' | 'setting' | 'arrival';

type GuidedSceneRow =
  | { kind: 'display'; label: string; value: string }
  | { kind: 'input'; label: string; placeholder: string; capturedKey: SceneInputKey };

The narrow SceneInputKey (instead of the spec's keyof CapturedInputs['scene']) keeps the TextInput binding type-safe — concepts: string[] and genre_response: string aren't surfaced as inputs in this card.

plan.ts — existing four rows tagged kind: 'display'; teaching mode appends Audience + Setting input rows; devotional appends What you bring. Quick and Deep are unchanged in row count.

StudySessionScreen.tsx

• Loads CapturedInputs once per session (getCapturedInputs) and rehydrates inputs (acceptance Merge pull request #3 from CraigBuckmaster/codex/review-recent-master… #4)
• Scene row render branches on row.kind — display keeps existing label/value pair; input renders a TextInput styled to match the existing observe input (bgSurface background, border border, body 14/21)
• Typing fires a 500ms-debounced setCapturedInputs (acceptance Merge pull request #2 from CraigBuckmaster/master #3); per-session timer reset on sessionId change; pending timer cleared on unmount

Tests

• 20 plan.parity snapshots refreshed and manually inspected:
  • Teaching fixtures (gen_1, isa_53, prov_3, psa_23, rom_8) → 4 display rows + Audience + Setting inputs ✓
  • Devotional fixtures → 4 display rows + What you bring input ✓
  • Quick + Deep → 4 display rows, no inputs ✓

Acceptance criteria

• Teaching shows Audience + Setting inputs; Devotional shows "What you bring"
• Quick + Deep show no extra input rows
• Typing persists via setCapturedInputs within ~500ms (debounced)
• Reopening the session rehydrates typed values (initial getCapturedInputs load)
• tsc / lint / full suite (3785 tests) clean

Rollback

Revert the PR. The captured-inputs columns from #1731 stay (nullable, append-only) and unused entries cause no issues.

Notes for Craig

• Visual: spec mentioned base.bgInput; the palette uses base.bgSurface for input backgrounds (matches the existing observe TextInput) so I used that for consistency. Easy to swap if you'd rather a dedicated token.
• Kanban move requires manual action.

---

Generated by Claude Code

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3785	❌ 0	3785
Suites	✅ 509	❌ 0	509

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 89.5s