Correct some alpha/stencil/blend issues in softgpu

[unknownbrackets]

This fixes a few issues found while implementing and validating hardware operations for #15163.

• Fixes stencil decrement on 5551 (decr means zero, not 0xFE swizzled to 1.)
• Uses zero for dst alpha on 565, though it'd be a bit odd for any game to actually use this.
• Cleanup warnings and force off a flag in PixelFuncID for invalid values.
• Corrects off-by-one errors in alpha blending.
• Returns vertexjit names for code pointers.

-[Unknown]

[hrydgard]

Cool.

I see, by the way, that rewriting all this to do 4 pixels at a time (in this pixel processing code code, probably pack 2 pixels in each register and using twice the registers, or go AVX2) would make everything way harder to work with. So maybe sometime in the future, but not while this is all still being worked out.

Also, if it helps to exclude 32-bit x86 from the softjit, to get more registers or whatever, then that seems fine to me.

[unknownbrackets]

Yeah, I think making it do two pixels at once would be easier, but masked writes (including stencil) are also an annoying area. That said, it's still something I want to pursue, I just have to make it smart...

-[Unknown]