[BUG] WebGPU detection

[fan-wen]

I've enabled WebGPU in my Chrome browser - Version: 132.0.6834.160 (Official Build) (arm64) by going to chrome://flags/#enable-webgpu-developer-features

Bug 1: I restarted the browser to make sure the feature is enabled. However, it seems like WebGPU couldn't be detected when loading the page (the MODnet option is not available).

Bug 2: After picking a sample image or uploading an image, the MODnet option becomes available. But it also becomes available when WebGPU feature is NOT turned on - should be a bug.

Bug 3: After picking the MODnet option (with the WebGPU flag enabled), it shows the following error:

process.ts:84 WebGPU initialization failed: Error: no available backend found. ERR: [webgpu] Error: WebAssembly is not initialized yet.
at async Promise.all (/index 0)
at async Promise.all (/index 0)
at async Gt (process.ts:73:19)

Bug 4: when processing an image with the default model, it shows NaN when loading model:

main.jsx
index.js:1 content script loaded
process.ts:144 Loading model: NaN%
process.ts:144 Loading model: NaN%
process.ts:144 Loading model: NaN%
process.ts:144 Loading model: NaN%
process.ts:142 Unknown model class "SegformerForSemanticSegmentation", attempting to construct from base class.
process.ts:142 Model type for 'SegformerForSemanticSegmentation' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.
process.ts:142 dtype not specified for "model". Using the default dtype (q8) for this device (wasm).
process.ts:144 Loading model: NaN%
process.ts:144 Loading model: NaN%
465process.ts:144 Loading model: NaN%
process.ts:144 Loading model: NaN%
process.ts:248 Processing images...
process.ts:255 Successfully processed image sample-image.jpg
process.ts:261 Processing images done