During some cursory profiling of impeller web, I notice that we call glCheckFramebufferStatus multiple times per frame. At least on WebGL, this is extremely slow, to the point where those calls alone are taking up 30+ms per frame (whereas the other rendering is only taking roughly 6 ms). According to MDN, glCheckFramebufferStatus can result in a full flush + round trip to the GPU.

We should consider only doing glCheckFramebufferStatus in debug builds, and simply assuming the framebuffer is complete in release builds.