fix(hyprland/language): split activelayout on the last comma outside parens

[vanshdev0101]

Closes #4586.

The bug

The activelayout payload is <keyboard>,<layout>, and both halves can contain commas — the keyboard in its vendor string, the layout inside its variant parentheses. Language::onEvent handled the latter by truncating the payload at the last ( and taking the last comma of what remained.

That assumes every bracket belongs to the layout. Keyboard names carry them too:

activelayout>>ite-tech.-inc.-ite-device(8910)-keyboard,Russian

truncates to ite-tech.-inc.-ite-device, which has no comma, so onEvent returned early and the layout never updated. The module goes blank on every switch for anyone whose keyboard has a bracket in its name.

Note this is independent of keyboard-name: the parse bails before that check runs, so setting it doesn't help.

The fix

Scan back for the last comma outside parentheses. That keeps commas inside a variant like English (US, intl., with dead keys) protected, while ignoring brackets anywhere in the keyboard name. The helper (rfindOutsideParens) went into util/string.hpp so it could be unit-tested without pulling in ALabel/IPC.

The issue suggests porting niri/language's logic. That doesn't transfer — niri's IPC returns a structured layout list (keyboardLayoutNames()) and never parses a payload, so there's no equivalent code to lift.

Testing

Reproduced on real hardware — a Bluetooth device that Hyprland registers as a keyboard with a bracketed name, on Hyprland 0.56.2. Given a second layout on that device only, switching layouts before the fix:

[debug] hyprland IPC received activelayout>>vis-s226-(avrcp),Russian
[debug] hyprland language didn't find matching layout
[debug] hyprland language formatted layout name LANG=[]

and after:

[debug] hyprland IPC received activelayout>>vis-s226-(avrcp),Russian
[debug] hyprland language onevent with Russian
[debug] hyprland language formatted layout name LANG=[Russian]

Added test/utils/string.cpp covering rfindOutsideParens and the payload split, including the reported payload and the variant-with-commas case the old ( heuristic existed to protect. I checked the tests fail against the previous algorithm rather than only passing against the new one — 3 of 11 assertions fail when the helper is reverted, including the #4586 case. meson test is green (waybar, hyprland, utils), full ninja build clean, clang-format clean.

util/string.hpp gained <algorithm>, <string_view> and <vector> includes: it was relying on transitive ones, and the new test is the first place that includes it standalone.

What I did not test: a physical keyboard whose name contains a bracket — the reproduction above uses a Bluetooth AVRCP device Hyprland exposes as a keyboard, which produces the same payload shape. I also only exercised this on Hyprland 0.56.2.