Complements isYUpInFramebuffer().

For instance the Qt Quick renderer will want the Batchable variant for
a vertex shader when rendering merged batches. Make it possible to
communicate this to the backends.

Plus some doc cleanups.

Focus on those that are used in combination with the pipeline state
and starting a renderpass.

beginFrame() has no applicable flags as of now but the flags parameter is
there for future use.

Sharable resources, which are at least buffers and textures, can now be
orphaned, meaning the owning QRhi can be destroyed before releasing the
QRhiResource. They will still function in a limited way (cannot be rebuilt
anymore) as long as there is at least one QRhi associated with the
QRhiResourceSharingHost (i.e. as long as the share group is non-empty).

The threaded case cannot be supported on all backends for now.
Revert parts of the previous d3d patch.

Some day we may consider exploring the possibility of doing something
with deferred contexts on d3d11 but that's not an option for now.

Here we choose to go the heavy weight route and turn it into a
d-pointered (implicitly shared, even) class since the data a binding
description needs may grow in yet-unknown ways when introducing support
for other resource types.

...and some leftover docs

...because we only really need to make sure arg changes are not skipped
upon a new pass, just changing the pipeline state has no effect on this.

Clarify the docs a bit as well.

Esp. for Metal where we now avoid superfluous set* for shader
resources.

There was a reason for defering (two passes update the same dyn.buffer
in the same frame, those can be accumulated instead of doing a full
DISCARD write twice) but the logic was broken due to optimizing out the
BindGraphicsPipeline in some cases. So keep it simple for now, the
multi-update in same frame case is discouraged anyway.

...also write more sophisticated output in the profiler
stream when it comes to frame timings.