x86/microVU: Reference StateEnd instead of inlining state

[stenzek]

Description of Changes

Instead of emitting a copy of the pipeline state at the end of the block's code, we can simply reference back to the block itself.

This will reduce icache pressure ever-so-slightly, since the pipeline state is 96 bytes.

I don't expect this to make a measurable difference on its own, but all the tiny things do add up.

Rationale behind Changes

Ongoing performance work.

Suggested Testing Steps

Check some games sensitive to M-bit, this previously wasn't using the fast copy routine. I think I've done this correctly, but don't know which games hit this path.

[refractionpcsx2]

Tried a few M-Bit games and a couple of games I had issues with state in the past when making it possible to exit the VU's and they all seemed fine.

I'll wait for the testers but I would say it LGTM