[js/webgpu] expose a few properties in WebGPU API

[fs-eire]

Description

This change exposes a few properties in ort.env.webgpu to resolve feature requirement mentioned in properties in #14579 (comment).

• Add powerPreference and forceFallbackAdapter in ort.env.webgpu, to allow users to set the value of the properties before the first inference session is created.
• Add readonly property adapter in ort.env.webgpu to allow users to get the adapter instance. Now users can access ort.env.webgpu.device and ort.env.webgpu.adapter.

@xenova @beaufortfrancois

[fs-eire]

So far, the design does not allow using different adapter or device for different inference session. Please let me know whether you have this requirement.

[xenova]

Great stuff! 🔥

So far, the design does not allow using different adapter or device for different inference session. Please let me know whether you have this requirement.

I don't think that will be necessary at the moment. Seems like a very niche use-case.

[beaufortfrancois]

LGTM

[xenova]

As a follow-up to this, do you think it makes sense to allow the user to set the adapter themselves? As stated in the comments of the code:

This property is only available after the first WebGPU inference session is created.

Which means I can't check for something fp16 support before creating the inference session. My use-case is for transformers.js to select the default device to use, depending on whether fp16 is supported or not. Of course, I can define a helper like this,

async function hasFp16() {
  try {
    const adapter = await navigator.gpu.requestAdapter()
    return adapter.features.has('shader-f16')
  } catch (e) {
    return false
  }
}

but this may not be the same device (especially if we update powerPreference and forceFallbackAdapter.