## What happened

On a system with a discrete NVIDIA GPU, the hardware-specific env vars that `default/hypr/nvidia.lua` is supposed to set (`NVD_BACKEND`, `LIBVA_DRIVER_NAME`, `__GLX_VENDOR_LIBRARY_NAME`) never actually reach the environment — not for Hyprland's own env, not for anything it launches, and not for the systemd/dbus activation environment that `omarchy-launch-shell` and other autostart apps get their env from.

## Root cause

`default/hypr/nvidia.lua` gates each `hl.env()` call on `o.shell_succeeds(...)` (in `default/hypr/helpers.lua`), which wraps `os.execute()`. Called from inside Hyprland's own Lua config, `os.execute()` cannot retrieve the exit status of the process it spawns — it always returns `(nil, "No child processes", 10)`, even for a trivially-true command like `os.execute("true")`. This is the classic `ECHILD`/"ignored SIGCHLD" symptom: Hyprland reaps/ignores `SIGCHLD` for its own children, so the `wait()` call inside Lua's `os.execute()` can never find the child it just forked.

Because `o.shell_succeeds()` always returns `false` in this context, every branch in `nvidia.lua` that depends on it (`omarchy-hw-nvidia`, `omarchy-hw-nvidia-gsp`, `omarchy-hw-nvidia-without-gsp`) is always skipped, regardless of what GPU is actually installed. The three env vars are silently never set.

I confirmed this is not specific to the nvidia detector scripts — a bare `os.execute("true")` called from a `require()`'d module inside `hyprland.lua` reproduces the same `nil, "No child processes", 10` result. By contrast, `io.open()` and `io.popen()` (for reading output, not exit status) work fine from the same context, so any Lua-config code that needs to shell out and check success/failure is affected, not just this one file.

## Steps to reproduce

1. On a machine with a discrete NVIDIA GPU, add this to `~/.config/hypr/hyprland.lua` (in a `require()`'d module, or inline) and `hyprctl reload`:
   ```lua
   local ok, kind, code = os.execute("true")
   hl.env("DEBUG_TRUE", tostring(ok) .. "|" .. tostring(kind) .. "|" .. tostring(code))
   ```
2. Spawn a fresh process via Hyprland and inspect its environment, e.g.:
   ```
   hyprctl dispatch "hl.dsp.exec_cmd(\"env > /tmp/envtest.txt\")"
   ```
3. `DEBUG_TRUE` in `/tmp/envtest.txt` reads `nil|No child processes|10` instead of `true|exit|0`.
4. Separately, confirm `NVD_BACKEND`, `LIBVA_DRIVER_NAME`, and `__GLX_VENDOR_LIBRARY_NAME` are absent from `systemctl --user show-environment` after a full session restart, even though `/usr/share/omarchy/bin/omarchy-hw-nvidia` and `omarchy-hw-nvidia-gsp` both exit 0 when run directly from a shell.

## Expected

`NVD_BACKEND`, `LIBVA_DRIVER_NAME`, and `__GLX_VENDOR_LIBRARY_NAME` should be set for any system with a detectable NVIDIA GPU, and should reach the systemd/dbus activation environment that autostart apps use.

## Workaround

Reimplemented the same PCI vendor/class/device detection in pure Lua using `io.open`/`io.popen` (reading sysfs files and directory listings directly) instead of shelling out and checking an exit code. That works correctly and sets the three vars as expected. Happy to turn this into a PR against `nvidia.lua` and/or `helpers.lua` if that approach (avoiding `os.execute()`'s exit-status path entirely) is the right general fix, rather than something Hyprland-side.

## System details

- Omarchy version: 4.0.0-1 (stable channel)
- Hyprland: 0.56.2, commit `efb50993780079460b0cbed1363e2166a2de1d9f`
- Kernel: 7.1.8-arch1-3
- GPU: NVIDIA GeForce RTX 5070 (GB205, Blackwell, well above the GSP device-ID threshold), alongside an AMD Ryzen "Granite Ridge" iGPU (`amdgpu`) — `omarchy hw hybrid gpu` reports this as a hybrid GPU configuration
- NVIDIA driver: 610.57.04, kernel modules `nvidia`, `nvidia_drm`, `nvidia_modeset`, `nvidia_uvm` all loaded and bound correctly
- Session: uwsm-managed Hyprland session (`uwsm start -g -1 -e -D Hyprland hyprland.desktop`)