OpcodeDecoding: Cache vertex sizes

[K0bin]

I struggled a bit to find a place where to put the cache but I'm pretty happy with having it in the callback.
The hit rate came out at 97% in Mario Galaxy. (I hooked it up to the stats HUD for testing.)

Together with #11066 Mario Galaxy goes from 85 fps to 140 fps on the hub world. (5900X, downclocked to 2.2 GHz)

[K0bin]

The new version just uses the vertex size that's stored in the VertexLoader if that's available.
Works for 99% of all calls in Mario Galaxy.

[Nerboruto]

could it bring some improvement in other games like fzero or metroid prime?

[K0bin]

I think @JMC47 tried it with those and it was between 2 and 5% faster.

[K0bin]

Ready for another round of reviews.

[iwubcode]

Untested but code wise LGTM. Great work @K0bin !!

[K0bin]

sigh

Either my local clang-format doesn't pick up the Dolphin config file or the CI bot has a different set of rules.

[Pokechu22]

Looks good to me. Most of my testing was on single core which didn't show much of an improvement (possibly because of other accuracy settings I have enabled), but I saw a lot larger of an improvement on dual-core. Thanks!

Regarding clang-format: I use MSVC's built-in formatting functionality, which picks up on the config file properly. I think there's also different behavior depending on the version of clang-format in use; I believe clang-format-9 is in use (and different versions behave differently with the same config).

[theofficialgman]

I've tested this and the preceeding PR
unfortunatly no FPS improvement seen here (nintendo switch, tegra-x1 quad core cortex A57 @ 2Ghz). 25FPS in that scene before and after the PRs

[badkarma12]

Small boost to Pokemon Colosseum/xd on scenes with shadow pokemon aura and moves like sunny day.