Fix SeparatePanel fullscreen origin on Windows

[anriltine]

Summary

• After showFullScreen(), explicitly move/resize the frameless SeparatePanel to the current screen geometry.
• Fixes Mini Apps that were resized for fullscreen but stayed anchored at the pre-fullscreen panel position on Windows (often with multiple monitors).

Context

Reported in telegramdesktop/tdesktop#30963 (also referenced by #31041 / #31114).

Root cause matches QTBUG-39537 / QTBUG-86899 for frameless windows: QWidget::showFullScreen() applies the fullscreen size but drops the move to the screen origin. Media viewer already works around this by setting screen geometry explicitly; Mini Apps use SeparatePanel, which previously only called plain showFullScreen().

Test plan

• Windows, single monitor: open a Mini App, enter fullscreen via UI or web_app_request_fullscreen — panel should cover the full screen from the monitor origin.
• Windows, two monitors: main Telegram window mid-primary; open Mini App fullscreen — panel should cover the primary screen, not remain centered over the main window and overflow.
• Exit fullscreen — panel returns to the normal anchored size/position.
• Linux/macOS smoke: Mini App fullscreen still enters/exits correctly.

Made with Cursor

[CLAassistant]

All committers have signed the CLA.

[ilya-fedin]

You're referencing a fixed Qt bug

[ilya-fedin]

Root cause matches QTBUG-39537 / QTBUG-86899 for frameless windows

QTBUG-39357 is fixed long time ago while QTBUG-86899 seem to be completely unrelated (it claims maximized windows behave like fullscreen ones on non-primary monitors and hide taskbar which doesn't seem to be the problem here)?

[ilya-fedin]

The root cause could be not a Qt bug at all but our bug: Qt positions windows automatically only unless setGeometry() or move() are called, RpWidget initializes with zero geometry. The Linux transient parent code path already works around that by setting WA_Moved attribute to false, maybe the same should be done before showFullscreen().

[anriltine]

Thanks, that matches what the Qt sources do.

QTBUG-39537 / QTBUG-86899 was the wrong diagnosis: current Qt already moves a fullscreen window to the monitor rect in QWindowsWindow::setWindowState_sys(). The remaining failure is on our side. RpWidget starts with setGeometry(0, 0, 0, 0), which sets WA_Moved, so QWidgetPrivate::create / QWidget::show() pass that position into the QWindow (setGeometry) instead of resize-only placement at the screen origin.

The previous patch also fought Qt: SetGeometryAndScreen() after showFullScreen() goes through setGeometry_sys(), which clears Qt::WindowFullScreen when not inside setWindowState().

Updated to the same workaround as the Linux transient-parent path in initGeometry(): setAttribute(Qt::WA_Moved, false) before showFullScreen().

[anriltine]

@ilya-fedin thanks again for the review. #342 was the wrong approach (QTBUG citations + SetGeometryAndScreen() after showFullScreen()).

Opened a clean PR with the WA_Moved fix you suggested, based on current master: #343

[anriltine]

Superseded by #343.