Remove pool2d MLRoundingType - Simplify the operand layout support of conv2d and pooling 2d operations

[fdwr]

Fixes #324

Rather than having two output shape parameters, an explicit output size and an output rounding mode, be more explicit in WebNN such that the caller passes the explicit output shape, resolving any higher-level policy beforehand (like with convTranspose2d):

dictionary MLPool2dOptions : MLOperatorOptions {
  sequence<[EnforceRange] unsigned long> windowDimensions;
  sequence<[EnforceRange] unsigned long> padding;
  sequence<[EnforceRange] unsigned long> strides;
  sequence<[EnforceRange] unsigned long> dilations;
  MLInputOperandLayout layout = "nchw";
- MLRoundingType roundingType = "floor";
  sequence<[EnforceRange] unsigned long> outputSizes;
};

Additional changes:

• Added spec:infra; type:dfn; text:string due to new build failures unrelated to my change (thanks Dominique).
• Rewrapped instanceNormalization for linter error.
• Update linter to display line numbers (save valuable time diagnosing errors) and display success message (rather than confusing silence).

---

Preview | Diff

[jlewis-austin]

The outputSizes validation above in 13.2 doesn't appear to be correct--shouldn't it be similar to this logic?

[fdwr]

I think this is now resolved by Joshua?

[inexorabletash]

This implicitly resolves #374 as well, doesn't it?

[inexorabletash]

This PR needs a redo since there's been a lot of spec churn since Nov 2024. But one thing I noticed in the current spec text:

If options.outputSizes exists, or if options.padding does not exist, set options.padding to the list « 0, 0, 0, 0 ».

That implies that if outputSizes becomes a required option, then any passed padding will always be ignored. Flaw in the current text? I think the first clause "if options.outputSizes exists" is bogus today, and should be removed.

[fdwr]

I'm confused by the build errors, as I didn't touch any of those lines, and [=string=] seems unrelated 🙃:

      LINE 1713:36: Multiple possible 'string' dfn refs.
      Arbitrarily chose https://infra.spec.whatwg.org/#string
      To auto-select one of the following refs, insert one of these lines into a <pre class=link-defaults> block:
      spec:infra; type:dfn; text:string
      spec:rdf12-concepts; type:dfn; text:string
      [=string=]

[dontcallmedom]

the build error is indeed unrelated to the particular change, but linked to the definition set bikeshed uses for auto-linking.

the simplest short term solution is to add spec:infra; type:dfn; text:string in the existing <pre class="link-defaults"> on L50.

[fdwr]

@dontcallmedom Thanks, that helped. I encountered a few other random errors (some which were unrelated to my change 🙃), and so I had to extend the linter to display line numbers to save my sanity:

index.html:14723: Unexpected normative reference to [WEBNN]

[huningxin]

From another perspective, would dropping roundingType prevent us from supporting dynamic shape (specifying the concrete input shape at the inference time) in the future? On native ML API side, both CoreML and ONNX supports ceil_mode which is same as PyTorch’s ceil mode.