zink: format issues
Not to be confused with the format regressions ticket.
We have an issue with 3-component RGB formats, specifically that sometimes we get these formats with either
PIPE_BIND_RENDER_TARGET when creating a resource, but then later gallium gives us the same resource with the other usage after it's been created, usually for blitting or somesuch, and then we can't use it because the resource was created after only checking for the initial usage with the vk driver.
Ideally I think we'd work around this by doing what other drivers do and always rejecting these 3 component formats so that gallium gives us a nice 4 component one, and I have a patch to do this in my branch. The issue is then that this causes some other tests to fail.
Good conflicting extension cases seem to be