cli/formats: let the screen commands reach bundles with no .rat child

The developer-logo splash declares its sprites directly and has no .rat layout
child, so is_build rejected it and no screen command could render it — despite
it being one of only two screens whose paint order has been read off the running
game, and the one where the layer key explains the whole permutation.

Adds ui_layout::is_composable (a declaration table plus at least one element
resolving to a T8aD the bundle carries) and an opt-in --all on screen
list/info/render. Measured on the disc: 2859 RATC bundles, 965 pass is_build,
2751 pass is_composable, and 0 pass is_build without passing it — a strict
superset. It is opt-in because the 1786 extra bundles are mostly two-element
fragments (a button and its glow), and because widening the default would
renumber --build for every pak, invalidating the build indices the corpus's
notes cite by number.

The splash now renders 6/7 elements, painting its glows first in the order
measured off the game; a disc test pins that order.

docs/re/structures/ui-composable-bundles.md

@@ -0,0 +1,86 @@
+# A screen build is not the only thing `compose` can draw
+
+**Status:** ✅ `CONFIRMED` by measurement over the disc, with the artifact to
+show for it. 🟡 the wider set is not "all screens" — most of what it adds are
+fragments. ❔ what distinguishes a screen from a fragment in the file is unknown.
+
+## The gap
+
+`ui_layout::is_build` — the predicate every screen command and every disc test
+enumerates with — requires a `.rat` layout child:
+
+```rust
+ratc::is_ratc(bundle) && kids.iter().any(|c| c.name.ends_with(".rat"))
+```
+
+The **developer-logo splash** has none. Its elements name their `T8aD` sprites
+directly (`palogo_gamearts.t32`), so there is nothing for a `.rat` record to
+place, and `parse_build` handles it perfectly well — it was simply never reached.
+
+That was worse than a missing screen. The splash is **one of only two screens
+whose paint order has been read off the running game**
+([`ui-paint-order-key.md`](ui-paint-order-key.md)), and it is the one where the
+layer key explains the *whole* permutation. Its measurement could not be checked
+against a render, because there was no way to produce one.
+
+## The measurement
+
+| | count |
+|---|---|
+| RATC bundles on the disc | 2 859 |
+| pass `is_build` | 965 |
+| pass `is_composable` (declaration table + a resolvable `T8aD`) | 2 751 |
+| passing `is_build` but **not** `is_composable` | **0** |
+
+So the new predicate is a strict superset, not a rival rule.
+
+**But the 1 786 extra bundles are not 1 786 screens.** They are dominated by
+two-element fragments — a button beside its glow:
+
+```
+GP_READY_ROOM.pak  +792     GP_MAIN_GAME_*2D.pak  +108 each (six languages)
+GP_DEBRIEFING_PILOTLOG.pak  +164     GP_HANGAR_ARSENAL.pak  +96
+  e.g.  2 els: ["pvbtnnew.t32", "pvbtnneweff.t32"]
+        2 els: ["py_menu_new.t32", "py_menu_new_eff.t32"]
+```
+
+That is why `is_build` stays the default. Widening it would also **renumber
+`--build`** for every pak, and the corpus's notes cite build indices by number
+(`GP_TITLE` build 4, entries 11/14, `GP_OPTIONS` build 11) — silently shifting
+them would invalidate written-down evidence.
+
+## What landed
+
+`ui_layout::is_composable`, and an opt-in `--all` on `screen list`, `screen info`
+and `screen render`. Default behaviour and default numbering are unchanged.
+
+```
+$ sylpheed-cli screen render dat/GP_TITLE.pak splash.png --all --build 11 --black
+build [11]: drew 6/7 elements
+  not drawn (1): ["palogo_eff0.prm"]
+```
+
+![the splash](../captures/ui-layout/developer-logo-splash-composed.png)
+
+All three logos with their glows behind them. The seventh element is a `.prm`
+primitive, which has no sprite and is skipped as everywhere else. A disc test
+pins the draw order to `[2,4,6,1,3,5]` — the glows first — which is the order
+measured off the running game, so the measurement is now checkable rather than
+merely recorded.
+
+Note this render also depends on the `_eff` fix
+([`ui-focus-and-effect-elements.md`](ui-focus-and-effect-elements.md)): before
+it, the three glows were dropped as "focused-state records" and the splash would
+have rendered as three bare logos.
+
+## What is not settled
+
+* ❔ **What makes a bundle a screen.** `is_composable` answers "can this be
+  drawn", not "is this a screen the game shows". Element count is a crude proxy
+  (fragments are 2–5 elements) and has not been checked against anything.
+* ❔ **The `.prm` primitives.** `palogo_eff0.prm` is a full-screen element with
+  pivot (640,360) and a single keyframe; nothing decodes `.prm` yet, so every
+  composite is missing whatever they draw. On the splash the measured paint order
+  puts it **first**, i.e. it is the backdrop.
+* 🟡 The splash render has not been diffed against a framebuffer capture. It is
+  now *possible* to, which it was not before.