feat(render): the two dead deliverable params are live — quiz count caps, mindmap orientation lays out (#63)

[danielPoloWork]

Problem (#63)

Two registry params were declared, validated, and then ignored — the manifest promised knobs that did nothing:

1. interview_quiz.count (short | standard | long) — build_quiz_ir never read params: count: short and count: long produced identical output.
2. mindmap.orientation (horizontal | vertical) — build_graph_ir ignored it and build_mindmap_svg was hardcoded horizontal.

The registry gate happily rejected count: shrt while accepting valid values that had no effect — validated-but-dead is worse than undeclared, and qbr-c-level.yaml itself advertises both knobs.

Fix — both params implemented (no third state)

• quiz count — build_quiz_ir resolves the registry params like the infographic/data projections already do, and caps the graded questions deterministically over the sorted ledger keys: short: 5, standard: 10, long: uncapped. The sorted cut makes the subset stable under manifest reordering; discussion questions are never capped.
• mindmap orientation — build_graph_ir stamps the resolved orientation into the graph-IR; build_mindmap_svg dispatches to the horizontal layout (unchanged — output verified byte-identical to the pre-change baseline) or the new _mindmap_vertical: root on top, one column per branch, leaves stacked beneath — deterministic like its horizontal twin (pure function of branch order, RFC-0002 §11-1). The shared _box helper is factored out of the closure.

Tests

• Cap per count value: 5 / 10 / 12 graded questions on a 12-cell ledger; discussion questions uncapped under short.
• Cap stability: reversing the manifest's inputs: order selects the same graded subset.
• Orientation: vertical changes the SVG bytes and stays deterministic; qbr-c-level (horizontal) stamps and renders as before.
• All reference manifests gate-green; 94/94 tests pass.

Closes #63

🤖 Generated with Claude Code