fix stencil resolve

It turns out, depth-stencil blits didn't work the last time I implemented them, because util_blitter only likes 1:1 blits. The only reason we didn't get an error is that we set the mask to R, which the destination didn't have, making the blitter conclude it was done before it did anything.

So here's a custom shader approach that makes us pass the n -> 1 sample blits. We still fail the 1 -> n sample blit-cases, because we can't write to multi-sampled stencil-buffers without PSSpecifiedStencilRefSupported. So that case still needs more work.