Chromium 131 validates depth bias on line and point topologies

[chrisdavidmills]

Summary

Chrome 128 deprecates setting depth bias for lines and points (Chrome issue, spec change).

This PR adds a datapoint for this change in behavior to GPUDevice.createRenderPipeline() and GPUDevice.createRenderPipelineAsync().

Test results and supporting details

See the description and data source

Related issues

See related MDN project issue: mdn/content#36341

[caugner]

@chrisdavidmills Are you aware of other spec changes that were documented in BCD in a similar fashion?

My two thoughts are:

1. This seems to be a rather small spec detail, and I'm not sure that we can maintain compat data at this level of detail (unless maybe if we find tracking bugs for the other engines).
2. Documenting it under createRenderPipeline() and createRenderPipelineAsync() does not seem ideal, as it looks like its part of the GPUDepthStencilState interface (which neither BCD nor MDN document yet).

PS: It looks like Chrome 131 removes the warning in favor of an error.

[chrisdavidmills]

@chrisdavidmills Are you aware of other spec changes that were documented in BCD in a similar fashion?

My two thoughts are:

1. This seems to be a rather small spec detail, and I'm not sure that we can maintain compat data at this level of detail (unless maybe if we find tracking bugs for the other engines).

2. Documenting it under `createRenderPipeline()` and `createRenderPipelineAsync()` does not seem ideal, as it looks like its part of the `GPUDepthStencilState` interface (which neither BCD nor MDN document yet).

PS: It looks like Chrome 131 removes the warning in favor of an error.

Ah right, so it looks like this should be Chrome 131 rather than 128 then.

So, we document GPUDepthStencilState on the createRenderPipeline() page. I really didn't want to have separate pages for all the different levels of WebGPU dictionaries, as it would make the docs way too complicated to understand. If we do end up keeping this, that is where the BCD should go.

But, if it is deemed to be too detailed, we could not include it as a separate data point.

[beaufortfrancois]

LGTM modulo question

[beaufortfrancois]

What is the deprecated field used for?

[chrisdavidmills]

If a feature has been officially deprecated, then we set "deprecated" to true. I know you used "deprecated" to describe what has happened, but when I looked at it, it seemed like a change in behavior rather than a true deprecation. Does that work, or do you think I should reconsider how I've described this?

[beaufortfrancois]

Spec said A, then they changed it to B.
We announced deprecation in Chrome 128 and removed it in Chrome 131.
I'll let you pick what is more appropriate in BCD.

[chrisdavidmills]

I think 131 makes more sense — when it starts to produce an error rather than the warning. I've updated it.