Float math in alpha calculation causes some punch-through textures to…

[flyinghead]

… have alpha noise.

[skmp]

I think the real bug is in https://github.com/reicast/reicast-emulator/blob/055c3599b6cfa99279542dc2d29c6e5032995ded/core/rend/gles/gles.cpp#L1713, should be /256.0f, not /255.0f

[skmp]

I think in general, we should be using 256 as 1.0 not 255

[flyinghead]

Isn't GL going to saturate to 1.0 for a value of 255 in an 8-bit channel? If that's the case then we should do the same.
I think the issue here is that we were using float math with >, which gives different results than integer math.

[AbandonedCart]

The added changes can be added later. As it stands now, there are some serious issues with alpha.