AI HTML recipe scrape of unsupported websites

[TomBursch]

Part of #670

[jmylchreest]

I work with an extraction solution that does a very similar thing to this, and I took the recent decision to actually build a small C based library and model registry - i'm just working on a small fine-tuned model based on Qwen at the moment - all of this will allow the project to leverage local inferrence within the app with a small (~200MB) local model, avoiding any inference costs etc and allowing users to do it all on-device. I'll keep you informed if it goes somewhere quickly :)

I mainly wanted to put a note here, simply because it has a lot of overlap.

My other solution deals with this from an API perspective in a similar way. Cleaning content prior to sending it for inference is quite important, and the refyne library provides that too. This isn't supposed to be an ad, just to say I have a brief demo app that does much the same thing here: https://recipeapp-demo.refyne.uk that I'd be happy to share/offer a key for refyne for if you're interested. The above has a re-implementation of the cleaner in rust to provide C bindings that the kitchenowl app/api could both use directly.