safety margin 1-2 ms on Windows, tested only on one machine (mine). no idea about linux. they use 3 ms
no need for safety margin in adaptive sync
if render journal starts logging really high numbers (bigger than the vblank interval) then there is no real point to RenderJournal anymore, it will cause problems

better to log peaks and times. not all of them
that log is not appropriate.
	do some exponential decay or something. with slower decay if there are multiple peaks

https://invent.kde.org/plasma/kwin/-/blob/master/src/renderloop.cpp
src/renderloop.cpp
	does the main work
src/renderjournal.cpp
	keeps a log of past frame times. but doesn't use fences, so it sees only the CPU side. I doubt the utility of this
	not sure how to do fences without either waiting on the main thread or creating a new thread


render policies don't make sense. use L2 time instead?
no need for smoothness though, which is what L2 optimizes for



does it know if there are no changes and hence doesn't need to render