Log input-to-scanout latency
If we can get the original input event timestamps from the kernel and they're in CLOCK_MONOTONIC
then we could diff that with the deadline timer (or generally the final flip timestamp) and report input-to-scanout latency. That would allow people to understand with exact numbers what the benefit of !2777 (merged) is. Although it would be more useful if we could apply the same logging to 44 and show how the latency was higher there.