CoreTiming: Improve frame pacing

[jordan-woyak]

@Sam-Belliveau took a step in the right direction with #11348 but throttling at every event was unnecessary.
They limited Throttle to 1200hz, but I think that ultimately lead to throttling on random events and not necessarily on critical presentation events.

See comment by @phire here:
#11348 (comment)

Ideally, you only want to throttle to make sure the Input/Output events. It doesn't matter if the emulator races though to complete rendering of half the frame as fast as possible, as long as it stops before reading input or presenting the result to screen (or audio, though we delay that)

Throttling is now done just before user input, statistics counting, and presentation.

GPU-thread cannot sanely use CoreTiming so a presentation_time is calculated from CPU-thread and sent in the "async request".
This works well for "single" and "dual" core mode. Ideally this time would probably be sent to the various video backends, since at least vulkan has its own worker thread, but for now it's handled in Present.

"Immediate XFB" does not gain this benefit. Timing information isn't readily available where that event is processed. I might argue that people wanting to use Immediate XFB care more about latency than pacing, though.

I'd also argue that needing to Throttle in PerfTrackerCallback is unfortunate and I plan to move this functionality into one of the other throttled events in a future PR to minimize the amount of throttling.

Frame pacing has become much better for me, from this:

To this:

Note the change from -/+ 0.22ms -> -/+ 0.02ms
I get an improvement like this in most games.

Input latency should, if anything, be better, since throttling happens just before input grabbing.

[iwubcode]

Code wise I think this looks good. I did test on Windows and didn't see the same buttery smoothness @jordan-woyak was getting. However, they quickly found a solution and now I get similar numbers.

Once there's some resolution from the Windows side, I think we can move forward here. Great work!

[iwubcode]

It was mentioned that we will save the Windows issues for a future PR. If so, this LGTM.

[JMC47]

I figure we should get this on the road then.