BufferData was always allocated in host visible mem, regardless of the
mem property argument.
NVidia gpus don't support blitting to 565 and 5551, which is needed for
mipmaps, so convert to 8888 instead.
Use optimal tiling for oit pointers image (linear not supported by
nvidia and probably slower anyway)
oit screen drawer uses multiple final attachments
Set non-oit depth attachment as transient and lazily allocate mem
fetch color format, storage buffer alignment and stores/atomics support
at init
support for framebuffers used as input attachments
use Allocator instead of VulkanAllocator
return handles instead of unique handles
Flyinghead