icd_shim, daemon: drop the shim's game-window report; clear the splash on the note

Merged
UT
18d ago
+23-429

With GameNative/gamescope#6 the compositor sends the game-window note itself on the first top-level X map, the same rule the X11 display server uses, so the shim's own reporting is dead weight: the present-path check every 120 presents, the X tree walk with the game-child predicate, the give-up after 240 walks, and the note sender in scanout.rs. All of it goes, along with the x11.rs helpers nothing else used. The GNW1 wire format in gndrm-uapi is unchanged.

The daemon no longer waits for the next published frame after the note: the first transition of the latch clears the splash directly from the scan-out thread, through the same permanent-attach upcall the commit worker uses for onFramePublished. The note now arrives at explorer's first map, before any game frame, and the app paints a black desktop for game sessions, so an immediate clear shows black either way. The per-presenter splash_cleared latch and its log line are removed with it. The daemon library ships in the app's jniLibs; the shim ships in the gamescope payload (20261023.4, built, not uploaded yet).

Verified on a Poco F6 with the app carrying this daemon and the payload carrying this shim: | Game (gamescope) | xwm: first top-level map | Splash hidden | Gap | First game frame |
|---|---|---|---|---|
| Balatro | +20.62 s | +20.64 s | black | +28 s |
| Blazing Chrome | +12.69 s | +12.72 s | black | +26 s |
| L.A. Noire | +21.71 s | +21.72 s | black | +52 s |

The hide now follows the first note within 20 ms; on the previous daemon it waited for the next published frame and in two runs for the once-a-second resend (1.9 to 3.2 s later). No gn-vkshim: x11 game-window predicate line appears any more, no blue frame, no taskbar, every game renders at its usual fps.