x11: fix dnd-kitten drag immediately cancelled by spurious synthetic ButtonRelease

[Copilot]

When the dnd kitten starts a drag, XGrabPointer is called from the tick callback after an async IPC round-trip — significantly later than the original Button1 press. Calling XGrabPointer(CurrentTime) at that point causes some X11 compositors to emit a synthetic ButtonRelease to transition from the implicit button-press grab to the new explicit grab. Since drag.active is already true and no XdndPosition target exists yet, handle_drag_button_release cancels the drag immediately. Tab dragging is unaffected because XGrabPointer is called synchronously within the same event cycle as the triggering motion event.

Changes

• glfw/x11_platform.h — add Time last_button1_press_time to the drag sub-struct
• glfw/x11_window.c
  • Record event->xbutton.time on every Button1 press
  • In _glfwPlatformStartDrag, pass the saved timestamp (falling back to CurrentTime if zero) to XGrabPointer instead of CurrentTime

Using the button-press timestamp is standard X11 DnD practice: it tells the server the grab was active from the moment the button was pressed, suppressing the spurious synthetic transition events regardless of how much wall time elapses before the explicit grab is requested.

[kovidgoyal]

@copilot that does not fix it. Doing further debugging, I find the release event is not coming from the X server but instead is coming from _glfwFreeDragSourceData in glfw/input.c