fix(aur): depend on virtual python-pywhispercpp instead of -cpu (#579)

The PKGBUILD hard-depends on python-pywhispercpp-cpu by name while
optdepends tells GPU users to "replace" it with the -cuda variant.
That swap is impossible in a normal pacman transaction: the -cuda
package conflicts with -cpu but only provides the virtual
python-pywhispercpp name, so removing -cpu breaks vocalinux's
dependency and pacman aborts. The only workaround is pacman -Rdd,
which leaves a broken dependency record that re-breaks on every
vocalinux upgrade.

All three AUR backend packages (-cpu, -cuda, -rocm) declare
provides=('python-pywhispercpp'), so depending on the virtual name
lets any variant satisfy the dependency. AUR helpers prompt for a
provider on fresh installs, the standard Arch pattern for
swappable backends.


Claude-Session: https://claude.ai/code/session_01XAF9BkcSzwcMjTzkBxdx6k

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

packaging/aur/vocalinux/PKGBUILD

@@ -19,7 +19,7 @@ depends=(
   'gobject-introspection'
   'python-cairo'
   'python-pynput'
-  'python-pywhispercpp-cpu'
+  'python-pywhispercpp'
   'python-pyaudio'
   'portaudio'
   'python-numpy'
@@ -42,7 +42,9 @@ makedepends=(
   'python-setuptools'
 )
 optdepends=(
-  'python-pywhispercpp-cuda: NVIDIA GPU (replace python-pywhispercpp-cpu)'
+  'python-pywhispercpp-cpu: CPU-only whisper.cpp backend (default provider)'
+  'python-pywhispercpp-cuda: NVIDIA GPU whisper.cpp backend'
+  'python-pywhispercpp-rocm: AMD GPU whisper.cpp backend'
   'python-onnxruntime: Silero VAD'
   'xdotool: X11 injection fallback'
 )