Validation test createBindGroup:buffer,resource_binding_size:* is unreliable #4660

The test contains the following TODO:

// MAINTENANCE_TODO: Allocating the max size seems likely to fail. Refactor test.

I have found this to be the case. When maxStorageBufferBindingSize is 2 GB, the test will create multiple ~2 GB buffers. On my Linux machine with 8 GB AMD GPU, it fails with OOM if I run the entire test, but passes if I run the three subtests (uniform, storage, read-only-storage) in separate invocations.

One strategy might be to split cases up further so that each storage type and test buffer size runs as a separate test with resource cleanup in between. But even then, it seems like a good test of this limit is necessarily fragile, since there's no guarantee that any particular request for a maximum-sized resource will succeed.

Also tracked by gfx-rs/wgpu#9642.