### What happened

`omarchy voxtype install` calls `voxtype setup gpu --enable` without `sudo`. That
command rewrites the `/usr/bin/voxtype` symlink, which needs root, so it fails:

```
✓ Setup complete!
Error: Failed to remove existing /usr/bin/voxtype (need sudo?): Permission denied (os error 13)
Try: sudo voxtype setup gpu --enable
```

The installer does not check the exit status (the call ends in `|| true`), so the
install reports success and finishes with the "Voxtype Dictation Ready"
notification. Dictation then works — but silently on the CPU build.

The failure is easy to miss: it is one line in the middle of a long install, and
nothing afterwards indicates the GPU step did not take effect.

### What was expected

On a machine where `omarchy-hw-vulkan` succeeds, the install should leave
`/usr/bin/voxtype` pointing at `voxtype-vulkan`.

### Steps to reproduce

1. On a machine with a Vulkan-capable GPU, run `omarchy voxtype install`
2. Accept the prompt and let it finish
3. `ls -la /usr/bin/voxtype`

Actual: `-> /usr/lib/voxtype/voxtype-avx2` (CPU)
Expected: `-> /usr/lib/voxtype/voxtype-vulkan`

### Cause

In `omarchy-voxtype-install`:

```bash
if omarchy-hw-vulkan; then
  voxtype setup gpu --enable || true
fi
```

The script runs as the user. `omarchy-pkg-add` elevates on its own, but this call
does not, so it cannot write to `/usr/bin`. Running `sudo voxtype setup gpu
--enable` afterwards works immediately:

```
Switched to GPU (Vulkan) backend.
```

and the daemon then loads onto the GPU:

```
whisper_backend_init_gpu: found GPU device 0: Vulkan0
whisper_model_load:      Vulkan0 total size =  1623.92 MB
INFO Model loaded in 0.74s
```

`nvidia-smi` confirms the VRAM is actually held (`/usr/bin/voxtype, 1564 MiB`).

### Suggested fix

Add `sudo` to that call. The script already declares
`# omarchy:requires-sudo=true`, so the credential is cached and this should not
prompt again:

```bash
sudo voxtype setup gpu --enable || true
```

Dropping the `|| true`, or warning when the step fails, would also stop a silent
CPU fallback from looking like a clean install.

### Impact

Any user with a Vulkan-capable GPU who installs dictation through Omarchy gets
the CPU backend without being told. On this machine the difference is
`voxtype-avx2` vs `voxtype-vulkan` — large-v3-turbo loads in 0.74s on the GPU and
is comfortably real-time; the CPU build is much slower for the same model.

### System

| | |
|---|---|
| Omarchy | 4.0.2-1 |
| voxtype-bin | 1.0.1-1 |
| CPU | 12th Gen Intel Core i7-12700F |
| GPU | NVIDIA GA102 [GeForce RTX 3080] |
| NVIDIA driver | 610.57.04 (open) |
| Kernel | 7.1.9-arch1-2 |

Happy to attach `omarchy debug` output if useful.