feat: added webgpu tensorflow example

[MendyBerger]

Adds a TensorFlow.js example running on top of wasi:webgpu

Needed before merging:

• next wash v2 rc (needs webgpu adapter info)
• replace wit-deps with wkg Version as rc.1 and add publish pipeline WebAssembly/wasi-gfx#63
• investigate a memory leak at the end of encoding
• investigate pretty major performance regression while refactor/cleanup
• slightly better ui with example image

Other external PRs that would allow removing some hacks here:

• feat(node): Filled in most util/types is* checks unjs/unenv#540
• Don't push array of one to features tensorflow/tfjs#8639
• chore: update wizer to support p2 bytecodealliance/ComponentizeJS#324

Other comments

• location.search is polyfiled to an empty string because tf assumes that's it's available.
• navigator.gpu is first set to empty an object at initialization so that the check if webgpu is available succeeds, and is later at runtime overridden with the real gpu object from wasi-gfx-js

[vados-cosmonic]

Hey @MendyBerger just double checking, but this is still in draft on purpose right?

[MendyBerger]

Yes, still need a few small changes. Will try to resolve soon

[ericgregory]

Follow-up README suggestions after local testing.

[ericgregory]

Consider adding a short Overview / How it works section between the Quickstart and the request instructions that tells the wasmCloud-specific story this example is built to demonstrate, including implementation details like what wasi_webgpu: true in .wash/config.yaml does.

[ericgregory]

Consider adding a Known limitations section before the "Adding Capabilities" link covering the rough edges a copy-paster will hit. Suggested content:

• Math.random is currently constant. polyfil.js overrides globalThis.Math.random to return 0.5 as a temporary shim until wizer initializes wasi:random. We should probably flag this.
• WIT fetching uses wit-deps instead of wash wit fetch. Mentioned above, but we should probably flag this explicitly.
• Resulting component.wasm is large. Might be worth noting and explaining why.

[MendyBerger]

@ericgregory really sorry to do such a large refactor after review. I did try to incorporate the changes you requested though.

This won't work without wasmCloud/wasmCloud#5122

[ericgregory]

@MendyBerger No worries, and this looks great! I only see a few small things after the refactor:

1. We should be as clear as possible in the readme about the wash version requirement - once the release with the fix exists, we'll want to point to that rather than wash 2.0 more generally, and we might want a quick "if you see X error, you need wash ≥ N/build off main" hint until then.
2. Are the nested component/wit/deps/ and service/wit/deps/ intended to be tracked? if not, could change the gitignore pattern to **/wit/deps/.
3. Signoff to pass DCO