VideoBackends:Vulkan: Improve backend multithreading

[TellowKrinkle]

Makes the multiple threads actually able to run at the same time

Fixes https://bugs.dolphin-emu.org/issues/12961

[AdmiralCurtiss]

Can someone review this?

[dvessel]

I never posted it but I did a quick test shortly after the PR was pushed and there was a slight improvement (this might be due to running a custom build with tello’s other PR’s) but it no longer makes a difference when toggling backend multithreading. There is a slight regression overall.

F-Zero - Sand Ocean intro 1x native

BEFORE (lastest master):

Multithreading enabled
1/1 vulkan..........................8.488ms

Multithreading disabled
1/1 vulkan.........................8.322ms

AFTER (merged with master):

Multithreading enabled
1/1 vulkan...........................8.812ms

Multithreading disabled
1/1 vulkan..........................8.670ms

Zelda - Wind Waker Intro 4x native

BEFORE (lastest master):

Multithreading enabled
1/1 vulkan.......10.042ms

Multithreading disabled
1/1 vulkan.......10.038ms

AFTER (merged with master):

Multithreading enabled
1/1 vulkan.......10.300ms

Multithreading disabled
1/1 vulkan........10.356ms

[TellowKrinkle]

Thinking about this more, we're waiting on the last submitted command buffer before submitting the next, which will hurt when GPU-bound.

We need to either fix this in a different way or increase the number of command buffers to 3

Will close this while I decide what to do about it

[TellowKrinkle]

Looks like all I had to do was wait
With #11090 merged, this is no longer an issue

Note: Skyward Sword was broken by #11090 on MoltenVK, fixed by #11122. Might want to hold out on testing this until it's merged. #11122 was merged

Seeing an improvement from ~58fps to ~64fps on Skyward Sword SC with #11122 and #10979 pre-applied, compared to Metal's ~66fps on an AMD Radeon Pro 5600M. On Intel graphics, which has a relatively slow Metal driver, MVK is now slightly faster than Metal at 58fps vs 55fps.

[TellowKrinkle]

Updated with a new, less hacky implementation
Also pictures
Before (thread timeline of one frame of the Skyward Sword file select screen on SC):

After (time scales are approximately equal at ~18.5ms for the whole view):