VideoCommon: flush vertex manager if components change

[Tilka]

DON'T MERGE, this PR is only to get a FifoCI comparison. I replaced the workaround with a proper fix.

degasus helped a lot with explaining and debugging this issue.

[degasus]

LGTM

[dolphin-emu-bot]

FifoCI detected that this change impacts graphical rendering. Here are the behavior differences detected by the system:

• chibi-robo-zfighting on ogl-lin-intel: diff
• ed-lighting on ogl-lin-intel: diff
• melee-lighting on ogl-lin-intel: diff
• milotic-texture on ogl-lin-intel: diff
• sadx-ui on ogl-lin-intel: diff
• soa-black on ogl-lin-intel: diff
• ssbm-pointsize on ogl-lin-intel: diff
• ztp-grass on ogl-lin-intel: diff
• zww-armos on ogl-lin-intel: diff
• zww-waves on ogl-lin-intel: diff
• milotic-texture on ogl-lin-mesa: diff
• milotic-texture on ogl-lin-nv: diff

_{^{automated-fifoci-reporter}}

[delroth]

Any reason to make shadergen depend on more global state vs. the previous way of passing the relevant state through args?

This doesn't seem to be justified and I see it as a regression.

[degasus]

@delroth This state could be passed by argument for SetShader(), but there is no (nice) way to do the same for VertexManager::Flush. Right now, only the vertex loader manager will know the last state, but almost every videocommon class might call VertexManager::Flush (which calls SetShader within the backends...). So we must still use a global. In my opinion, it is cleaner to fetch the global where it's needed (shadergen), not within a completely different subproject (videocommon vs backends). If we want to pass the "state" by argument, we should think about a united "GPU state"... But this is out of scope, so just use the same as all other states else sounds fine for me.

[delroth]

Agreed. Thanks for explaining why the change is necessary (VertexManager::Flush).