Now that we no longer have a buffer border, we can render directly to
the video memory if the output surface is 8-bit and double buffering is
enabled.
Also adds a flag to force this even when double buffering is disabled,
because some systems emulate 8-bit mode with something else, so the
output is always buffered.
This required a tweak to the cursor to make sure that we clear it after
`SDL_Flip`.
Gleb Mazovetskiy
cb9da2cadc