Inhibit idle for fullscreen Steam games, not just the Steam client

[RandomJerk]

Closes the gap described in #11271.

Problem

default/hypr/apps/steam.lua sets idle_inhibit = "fullscreen" for class steam — the Steam client / Big Picture. Games don't run under that class; they run as steam_app_<appid>, so the rule never applies to the window that's actually fullscreen during play.

Gamepad input goes from Steam Input straight to the game via SDL and never passes through the compositor, so a fullscreen game played on a controller looks completely idle to Hyprland. idle.screensaver fires over the match at 150 s, idle.lock at 300 s. Steam's own inhibit and gamemode's inhibit_screensaver don't help — both use the org.freedesktop.ScreenSaver D-Bus interface, which Hyprland doesn't implement.

Change

One rule added next to the existing client rule:

o.window("steam_app_.*", { idle_inhibit = "fullscreen" })

• Separate line rather than widening the existing one, because that one also sets float = true, which games must not get.
• Unanchored pattern to match the file's existing idiom ("steam.*" on the opacity line). Hyprland matches rule regexes against the whole class, so steam_app_.* is a full match; verified against a throwaway window that the unanchored form does produce an inhibitor.
• Scoped to fullscreen so a windowed or alt-tabbed game returns to normal idle behaviour.
• The idle service already honours inhibitors (respectInhibitors: true in shell/plugins/services/idle/Service.qml), so nothing else changes.

Verification

Omarchy 4.0.3-1, Hyprland 0.56.2, NVIDIA 610.57.04. Rule applied, Proton/XWayland game relaunched, hyprctl -j clients polled:

23:13:35  steam_app_3468650  fullscreen=0  inhibitingIdle=False   (window mapped)
23:13:37  steam_app_3468650  fullscreen=2  inhibitingIdle=True    (fullscreen → inhibitor on)
23:30:19  (game closed)

Seventeen minutes on controller only, no screensaver. Without the rule the same session produced the screensaver at 150 s with inhibitingIdle=False.

Tests

./test/all: 231 of 236 test files pass. The five that fail (config, locate, snapper, unowned-system-paths, update-lock under test/shell.d/) fail identically on untouched quattro in this environment, so they're pre-existing and unrelated to this change.

🤖 Generated with Claude Code

https://claude.ai/code/session_01U1r4YEdkmTPPJUYHB5eqTG