Support running scripts with Nodejs or Deno, including webgpu support

Lets make it possible to test WebGPU accelerated code via scripts, e.g. when running: https://github.com/PAIR-code/tiny-transformers/blob/main/animated-transformer/src/lib/seqtasks/tiny_worlds_train_self_att.script.ts

This may help us debug memory leaks too.

Breadcrumbs:

General TFJS platform guide: https://www.tensorflow.org/js/guide/platform_environment

WebGPU support for tfjs:

• https://github.com/tensorflow/tfjs/tree/master/tfjs-backend-webgpu

TFJS Node support (references GPU support too, but for linux and windows, not yet mac)

• https://github.com/tensorflow/tfjs/blob/master/tfjs-node/README.md

Key issues for Deno (looks resolved... uses tfjs-backend-webgpu and seems to use Chrome GPU backend support):

• https://github.com/denoland/deno/issues/25874 : WebGPU adapter.info call fails after device creation [closed]
• https://github.com/denoland/deno/issues/28135 : GPUAdapter invalidation causes TensorFlow.js WebGPU backend to fail — Follow-up to [closed, but some discussion on if it can already be used...]

Possible alternative paths... webgpu for node and then make tfjs-node-gpu

• https://www.npmjs.com/package/webgpu & discussion about it in nodejs forum: https://github.com/orgs/nodejs/discussions/41994

Start with animated-transformer/src/lib/seqtasks/tiny_worlds_train_self.script.ts and the linked TensorFlow.js platform guide. Compare the documented Node.js and Deno paths, including the referenced WebGPU backend and runtime issues. Done means the script can be run for WebGPU-accelerated testing in the supported runtime(s), with memory-leak debugging possible.