[MacOS] example/validation bugs on M2

[Foxtr0t1337]

validate_new() always fail when using default BindlessContextCreateInfo

vulkano/src/descriptor_set/layout.rs:62
the total number of descriptors across all bindings is greater than the `max_per_set_descriptors` limit, and `device.descriptor_set_layout_support` returned `None`

According to my tests, on my MBP M2, total number is 3000100 (from the default BindlessContextCreateInfo) and max_per_set_descriptors is 1024 (from device properties).
I haven't investigated the device.descriptor_set_layout_support part but if I just new_unchecked() the examples work fine.

This happens in async-update, bloom, deferred because they use Resources::new(). ray-tracing also used the defaults but I can't test it.

Another one

examples/bloom/main.rs:507:10
Caused by:
    this device is a portability subset device, and `create_info.format` does not have the same components and number of bits per component as `image.format()`

Requires one of:
    device feature `image_view_format_reinterpretation`

After enabling that feature it works fine.

[marc0246]

This is unfortunately another case of spec ambiguity. It says that the max_per_set_descriptors limit applies to all descriptor sets; however, all other such limits only apply to non-update-after-bind descriptors. If the validation layers implemented validation for this limit, they would error out the same way because they also go by what the spec says. So spec issue it is.

[marc0246]

Actually, before opening a spec issue, I opened a MoltenVK issue because I think it's more likely their issue: KhronosGroup/MoltenVK#2573