wgpu-core should enforce proper nesting of push_debug_group and pop_debug_group

[jimblandy]

There is no code in wgpu-core or wgpu-hal to prevent unmatched push and pop calls at the wgpu or WebGPU level turning into unmatched Begin and End calls at the Vulkan API. But on Vulkan, VUID-vkCmdEndDebugUtilsLabelEXT-commandBuffer-01912 requires:

There must be an outstanding vkCmdBeginDebugUtilsLabelEXT command prior to the vkCmdEndDebugUtilsLabelEXT on the queue that commandBuffer is submitted to

After much discussion, the VulkanValidationLayers people seemed to decide that this was impossible to check in the validation layer, and the check was removed altogether. But the VUID is still in the spec, so Vulkan implementations have the right to assume the stack won't underflow.

wgpu-core should require that the command buffers submitted to a queue produce properly nested push and pop pairs, to ensure that our usage respects the specification.

Issue #3377 shows that even stricter limitations may be necessary to avoid bugs in certain versions of Mesa. But wgpu-core should always respect the Vulkan spec's general rules.

[teoxoy]

It's required by the WebGPU spec for these to be well-balanced, so we can implement the validation in wgpu-core.