Right now, there are two cases in the Blink C++
that use IntersectionObserver, but don't register via the same mechanisms as script.

1. HTMLMediaElement

https://cs.chromium.org/chromium/src/third_party/blink/renderer/core/html/media/html_media_element.cc?l=4269

2. MediaCustomControlsFullscreenDetector

https://cs.chromium.org/chromium/src/third_party/blink/renderer/core/html/media/media_custom_controls_fullscreen_detector.cc?l=141

This has two problems:

1. The call to ComputeGeometry happens at the wrong time. Instead of happening right after paint
during the rendering lifecycle, it happens whenever the event happens to run in the event loop

2. Extra tasks being inserted into the event loop

It may be that these call sites do this on purpose because they want a lower update rate
than regular script uses of IntersectionObserver. In both cases they only run a measurement
once a second.

https://cs.chromium.org/chromium/src/third_party/blink/renderer/core/html/media/html_media_element.cc?l=134
https://cs.chromium.org/chromium/src/third_party/blink/renderer/core/html/media/media_custom_controls_fullscreen_detector.cc?l=19

This seems to point to a hole in the IntersectionObserver script API, as developers probably
want the ability to throttle work for performance.

Also filed: https://github.com/w3c/IntersectionObserver/issues/298