GRAPHICS: Support having transparent ManagedSurface

[criezy]

A change made yesterday forces ManagedSurface to be fully opaque. This may be an issue for some engines.
This commit revert this and also adds proper handling of the surface alpha when blitting pixels.

[OMGPizzaGuy]

ManagedSurface::blitFromInner also blends, but it I think it's safe to assume it should always end with 0xFF as alpha there.

[criezy]

ManagedSurface::blitFromInner also blends, but it I think it's safe to assume it should always end with 0xFF as alpha there.

I think it might be better to also handle the destination transparency there. I will have a look.

[criezy]

I had a quick look at ManagedSurface::blitFromInner and at the very least it is not consistent between the case where the source and destination formats are the same and the generic case. In the former case it just copies the source onto the destination with no blending, which is probably incorrect if the source is partially transparent. So I think I will need to do some changes here anyway.

Edit: And the blending is also wrong. It doesn't blend with the destination pixel, but with the previous blended pixel...

[criezy]

I have now fixed the obviously wrong alpha blending in ManagedSurface::blitFromInner and decided to add support for transparent destination there.

I have played through a few games in three different engines that use ManagedSurface (AGS, Glk, and Cryomni3d) and did not see any regression.

[criezy]

Thank you for the review and for pointing out my stupid mistakes.
I will merge this now and cross my fingers that it does not introduce regressions..