Figma frame selection: consume dev-status (sectionStatus/handoffStatus) from Kiwi decode as primary signal, heuristics as fallback

## Signal (verified)

Figma's Dev Mode "Ready for dev" / "Completed" status is carried in the `.fig` Kiwi data, under internal field names (NOT the public REST `devStatus`). Verified by decoding the raw-deflate **schema chunk** of two real fixtures (`wp-cloud-2.fig`, `david-perell.fig`, schema version 106) — bounded read, no message decode:

- **`sectionStatus` / `SectionStatusInfo`** — the Ready-for-dev status, attached to **sections**
- **`handoffStatus` / `HandoffStatusMap` / `HandoffStatusMapEntry`** — dev-handoff state
- enum values: **`DEV_HANDOFF`, `DEVELOPMENT`, `COMPLETED`**
- related: `NodeStatusChange`, `currentStatus`, `statusInfo`

The REST/Plugin `devStatus` is a public projection of these. So we can read the signal from the Kiwi message we already decode — no REST dependency. This is the scalable path: schema-driven decode works for any `.fig`.

## Why this matters

Frame selection today is 100% heuristic — `device_hint` inferred from name+width, sibling grouping by name similarity, `max_pages`. That's pattern-based guessing at designer intent. `sectionStatus`/`handoffStatus` IS designer intent, stated explicitly. Using it:
- Selects/prioritizes deterministically — marked-ready sections in, WIP/drafts out.
- Kills the duplicate-draft false-positive (e.g. WP.Cloud's 4× "For Hosts" desktop drafts) with real data instead of width/name heuristics.
- Gives a natural page/component unit: a Ready-for-dev **section**.

## Scope

1. **Decode (the valuable part):** extend the Kiwi field policy so the existing schema-driven decoder carries `sectionStatus`/`sectionStatusInfo`/`handoffStatus` through into the scenegraph. No hand-rolled byte parsing — use the real Kiwi decode.
2. **Normalize:** map the internal enum (`DEV_HANDOFF`/`DEVELOPMENT`/`COMPLETED`) to a clean `dev_status` on nodes (`ready_for_dev` | `completed` | null), carrying the raw value alongside for auditability.
3. **Select:** when any `dev_status` is present, make it the PRIMARY frame-selection signal (prefer ready/completed sections+frames); demote the name/width heuristics to fallback used only when no dev-status exists in the file.
4. **Diagnose:** dev-status coverage (how many sections/frames carry a status, and the selection source: dev_status vs heuristic).

## Acceptance

- A scenegraph with a `Ready for dev` / `Completed` section selects those frames and skips WIP; with no dev-status present, behavior falls back to today's heuristics unchanged.
- Diagnostic reports dev-status coverage and which signal drove selection.

## Validation

Schema presence is proven. Value-level proof (a section actually marked ready) needs a **marked** fixture decoded — the FSE Pilot file has the label visible in-app and should be re-exported into the fixtures dir, then validated on the lab (not locally — large `.fig` message decode is RAM-heavy).

Start by locating the existing Kiwi schema-driven decoder and its field policy, then inspect the scenegraph frame-selection entry point and the fixtures directory. Validate the marked FSE Pilot fixture on the lab, checking that sectionStatus/handoffStatus values reach the scenegraph and that diagnostics report coverage and the selection source. Done means status-driven selection works while files without dev-status retain the existing heuristic behavior.

Written by the indexing model from the issue text.