## What happened

On a muxless hybrid laptop where the panel is wired to the **AMD iGPU** and the NVIDIA dGPU has **no connected outputs**, Omarchy sets `LIBVA_DRIVER_NAME=nvidia`. This silently breaks all hardware video decode in Chromium/Electron apps: **audio plays but video never renders**, and tabs visibly flicker.

It is not a soft failure. Because `vaInitialize` *succeeds*, Chromium never falls back to software decode, so there is no recovery path and nothing useful in the logs.

## Root cause

`default/hypr/nvidia.lua` gates on the mere *presence* of a GSP-capable NVIDIA GPU:

```lua
if o.shell_succeeds(o.shell_quote(nvidia)) then
  if o.shell_succeeds(o.shell_quote(nvidia_gsp)) then
    hl.env("NVD_BACKEND", "direct")
    hl.env("LIBVA_DRIVER_NAME", "nvidia")
    hl.env("__GLX_VENDOR_LIBRARY_NAME", "nvidia")
```

It never checks **which GPU drives the display**. On this machine only `card2-eDP-2` (amdgpu) is connected; every NVIDIA connector is disconnected and `nvidia-smi` reports `display_active: Disabled`.

Chromium deliberately refuses NVIDIA render nodes for VA-API:

```
WARNING:media/gpu/vaapi/vaapi_wrapper.cc:130] Should skip nVidia device named: nvidia-drm
```

so it decodes on `/dev/dri/renderD129` (amdgpu) — while `LIBVA_DRIVER_NAME=nvidia` forces libva to load the NVDEC driver there. With `NVD_BACKEND=direct`, nvidia-vaapi-driver reaches the dGPU over CUDA and **ignores the DRM fd it was handed**, so initialisation reports success against the wrong GPU. Decoded frames land in NVIDIA memory that the AMD-driven display cannot import.

Demonstrated by initialising libva directly against the AMD render node:

```
AMD node (renderD129) + LIBVA_DRIVER_NAME=nvidia
  OK  va 1.24  vendor=VA-API NVDEC driver [direct backend]      <-- wrong GPU, still "succeeds"

AMD node (renderD129) + LIBVA_DRIVER_NAME=radeonsi
  OK  va 1.24  vendor=Mesa Gallium driver 26.2.2-arch1.1 for AMD Radeon 860M Graphics (radeonsi, krackan1)

AMD node (renderD129) + LIBVA_DRIVER_NAME unset (autodetect)
  OK  va 1.24  vendor=Mesa Gallium driver 26.2.2-arch1.1 for AMD Radeon 860M Graphics (radeonsi, krackan1)
```

libva's own autodetection picks the correct driver. Omarchy's override is what breaks it.

## Steps to reproduce

1. Use a muxless hybrid laptop whose internal panel is driven by the AMD/Intel iGPU (NVIDIA dGPU has no connected connector).
2. Boot Omarchy; `default/hypr/nvidia.lua` exports `LIBVA_DRIVER_NAME=nvidia`.
3. Open any YouTube video in Chromium.
4. Audio plays; the video area stays blank and the tab flickers.

## Expected

Hardware decode works, or at minimum falls back to software decode rather than failing silently against a GPU that cannot present.

## Suggested fix

Only claim the NVIDIA VA-API driver when NVIDIA actually drives a connected output. Everything else in the block (`__GLX_VENDOR_LIBRARY_NAME`, PRIME offload) can stay as-is — it is specifically `LIBVA_DRIVER_NAME`/`NVD_BACKEND` that must follow the display, since VA-API surfaces have to be importable by whichever GPU scans out.

A sysfs-only detector, consistent with the existing "don't shell out to lspci" note in that file:

```sh
# omarchy-hw-nvidia-drives-display
for conn in /sys/class/drm/card*-*/status; do
  [ "$(cat "$conn" 2>/dev/null)" = "connected" ] || continue
  card="${conn%-*}"; card="${card%/status}"
  drv=$(basename "$(readlink -f "$(dirname "$conn")/../device/driver")" 2>/dev/null)
  [ "$drv" = "nvidia" ] && exit 0
done
exit 1
```

then in `nvidia.lua`:

```lua
if o.shell_succeeds(o.shell_quote(nvidia_gsp)) and o.shell_succeeds(o.shell_quote(nvidia_drives_display)) then
  hl.env("NVD_BACKEND", "direct")
  hl.env("LIBVA_DRIVER_NAME", "nvidia")
end
hl.env("__GLX_VENDOR_LIBRARY_NAME", "nvidia")
```

Simply leaving `LIBVA_DRIVER_NAME` unset on iGPU-driven systems would also be correct — libva autodetects `radeonsi` properly, as shown above.

Note this must also hold for hotplug: an external monitor on an NVIDIA-wired HDMI/DP port changes the answer at runtime. Gating on boot-time state is still a strict improvement over gating on nothing.

## Workaround

Append to `~/.config/hypr/hyprland.lua` (loads after the defaults):

```lua
hl.env("LIBVA_DRIVER_NAME", "radeonsi")
```

## System details

- **Omarchy:** 4.0.4-1
- **CPU:** AMD Ryzen AI 7 350 w/ Radeon 860M
- **GPUs:** NVIDIA GB206M RTX 5070 Max-Q `[10de:2d18]` (card1, renderD128) + AMD Krackan Radeon 860M `[1002:1114]` (card2, renderD129)
- **Connected outputs:** `card2-eDP-2` only — all NVIDIA connectors disconnected
- **Kernel:** 7.2.5-3-omarchy
- **Hyprland:** 0.56.2 / Aquamarine 0.15.0
- **Driver:** nvidia-open-dkms 610.57.04, mesa 26.2.2, libva 2.24.1, libva-nvidia-driver 0.0.18
- **Chromium:** 152.0.7977.82