Validate `maxComputeWorkgroupSize*` and `maxComputeWorkgroupStorageSize`

[andyleiserson]

CTS tests:

• webgpu:api,validation,compute_pipeline:limits,workgroup_storage_size:*
• webgpu:api,validation,compute_pipeline:overrides,workgroup_size,limits,workgroup_storage_size:*
• webgpu:api,validation,compute_pipeline:overrides,workgroup_size,limits:*

Note that they need to be validated at pipeline creation, after overrides have been resolved.

[teoxoy]

We have some validation, is it not working? #7429 (comment)

[ErichDonGubler]

@teoxoy's link discusses maxInvocationsPerWorkgroup, which is different from maxComputeWorkgroupSize* limits. I think his point still stands, though: we do seem validate these limits in compute shaders at wgpu-core/src/validation.rs:1393-1402. They also appear to be validated on render pipeline creation. It does seem we don't validate maxComputeWorkgroupStorageSize, though.

@andyleiserson, do you have definite evidence we aren't properly validating maxComputeWorkgroupSize* limits, or should we narrow the scope of this issue to mention maxComputeWorkgroupStorageSize only?

[andyleiserson]

It was rather useless of me not to include the selectors when I filed the issue, sorry. 😞

The issue with the invocation size limits is that they need to be validated after overrides have been resolved. We resolve overrides within the hal create_compute_pipeline function, but we validate the size in core's create_compute_pipeline function before calling hal.

Besides the order that things happen, we also need to be careful that we validate against the final, override-resolved version of the entry point, rather than using shader_module.interface.entry_points if that can be stale.