Support for @xenova/transformers seq2seq pipelines and webLLM

[MiguelsPizza]

The transformers.js api would provide a better UX for downloading and running local models. Inference could be done in the extension service worker which now supports webGPU