RFC: etnaviv: Set blit box sizes based on real sizes

The padded sizes are based on the real sizes multiplied with the samples per direction. Later, in etna_try_rs_blit, the box sizes are multiplied by the number of directional samples, so that the assertions fails for multisample blits.

Since the box size later converted to source samples anyway, I assume that the box size should be the logical size of the region. Consequently, the first patch avoids that the assertions in etna_try_rs_blit are hit when an MSAA blit is requested.

The second patch disables downsampling when the blit target is also an MSAA surface.

However the two patched don't really fix the one example dinoshade I was looking at, the first patch only makes sure the assertions do not fire.

All this was tested on GC2000.