Multisampling with multiview rendering

How would MSAA work? The resolveTexture of a color attachment can only be a (given mip level of a) 2D texture or (a given mip level of) a single layer of a 2D texture array.

QRhiRenderBuffer is out, clearly, but what if one uses a multisample texture array?
Do we just say the auto-resolve (resolveTexture) cannot be used then, and it is left to the application's shaders to resolve when using the result (sampler2DMSArray + texelFetch) ?
Would this work for Quick 3D? (not sure)

Or can we do something for the automatic resolve? (what does Vulkan resolve attachments support?)

In any case, this needs to be proven in a manual test, plus update docs if not obvious how this is done.

Update: proper handling of multisample texture arrays incl. resolve is doable for GL/Vulkan/D3D12. It just needs some fixes here and there.