Summary

Repeated fullscreen toggles on Qt6/Wayland can re-enter mouse capture while the pointer is already locked, which ends in the Wayland protocol error about the surface already being constrained.

This crashes 86Box so badly on my machine that 86Box.cfg is overwritten with defaults for some reason.

It started after 7ea6e9c began capturing the mouse on every fullscreen entry.
The Qt handler behind setMouseCapture was not idempotent, so duplicate requests still ran the backend capture path.

Fix it by treating mouse capture as a real state transition and ignoring requests that do not change the stored capture state.

Checklist

• Closes N/A
• I have tested my changes locally and validated that the functionality works as intended
• I have discussed this with core contributors already
• This pull request doesn't require changes to the ROM set
• This pull request doesn't require changes to the asset set