Single frame of garbage data when starting Gfx with a vram-backed framebuffer

[TapGhoul]

As per https://discord.com/channels/1410429555188961342/1410435424249450610/1410993383987744838

The general issue here is that when creating a framebuffer via https://github.com/devkitPro/libctru/blob/e09a49a08fa469bc08fb62e9d29bfe6407c0232a/libctru/source/gfx.c#L125-L158 we immediately present the framebuffer and turn of GSGPU's "force black" setting. When doing this with a CPU-side buffer, this is fine, as it's initialized to all 0's, but when it's a GPU-side buffer it can have anything in it. And until you present again, this remains on screen. As a result, you are forced to display a single frame of garbage on both displays.

A few proposed solutions:

• Call GSPGPU_SetLcdForceBlack(0x1) right after init, blanking out the display again, so we can clear it properly and call present again ourselves, burning a single frame of startup time and performing redundant startup work.
• Submit a patch to libctru to fix it.
• Implement the startup routine ourselves, with a fork of libctru exposing the required statics.
• Compile libctru within ctru-sys, allowing us to apply patches prior to a build.

[TapGhoul]

For the first option, this code seems to work well enough to prove the concept:

unsafe { assert_eq!(ctru_sys::GSPGPU_SetLcdForceBlack(1), 0) };
for _ in 0..2 {
    unsafe { assert!(citro3d_sys::C3D_FrameBegin(0)) };
    for target in [
        &mut bottom_target,
        &mut top_left_target,
        &mut top_right_target,
    ] {
        target.clear(ClearFlags::ALL, 0, 0);
        instance.select_render_target(target).unwrap();
    }
    unsafe { citro3d_sys::C3D_FrameEnd(0) };
}
gfx.wait_for_vblank();
unsafe { assert_eq!(ctru_sys::GSPGPU_SetLcdForceBlack(0), 0) };

The reason for passing in a flag of 0 to frame begin is that we can blit both buffers within one vblank, and then we manually wait for vblank prior to revealing the LCD again.
This does cause a momentary backlight flash, but the screen is still black while this occurs.

More ideal would be an explicit copy + swap, but as a proof of concept it works.