Screencasts invoked by RecordVirtual() do not respect cursor parameter.

Tested version: mutter/main, g-s/main, but it is very likely that all versions are affected since the introduction of RecordVirtual() in GNOME 40.

Reproducible: always

g-r-d specifies in screencasts the META_SCREEN_CAST_CURSOR_MODE_METADATA setting. With this setting, screencasts do not include the cursor in frames. Instead, the frame data + the cursor data are received in two distinct buffer, allowing to set the cursor image on the client side to propagate a smooth experience even if it wouldn't be (also no cursor lags due to latency).

This works fine with RecordMonitor(), but with RecordVirtual() mutter always blits the cursor image ontop of all frames, regardless of the cursor setting.

This is especially visible with mstsc as client:

As you can see in the screenshot, the cursor in mstsc did itself not change. This means, that g-r-d never received a PipeWire buffer containing cursor data.
What can be seen too in this screenshot is that the cursor is painted on top of the frame.

CC: @jadahl