Update conversational inference API snippets

[mishig25]

Description

This PR updates inference snippet generating functions signatures. Before, the functions were generating string. Now, the function will generate InferenceSnippet | InferenceSnippet [].

huggingface.js/packages/tasks/src/snippets/types.ts

Lines 14 to 17 in 5bc694b

	export interface InferenceSnippet {
	content: string;
	client?: string; // for instance: `client` could be `huggingface_hub` or `openai` client for Python snippets
	}

Why do we need to generate InferenceSnippet [], not just InferenceSnippet?
Because for a given langauge (let's say), we wanna show multiple clients options (huggingface_hub, openai). (see the attached video below).

Also, this PR improves the conversational snippet greatly.

Screen recording

Screen.Recording.2024-10-22.at.11.32.06.mov

[mishig25]

ready for review

[Wauplin]

Thanks @mishig25! Results looks amazing and will definitely settle the debate over on which snippets to display 😄 As mentioned on the moon-landing PR, I would tend to always return InferenceSnippet [] to avoid confusions but if you think otherwise, I'm really fine keeping as it is now.

Once this is merged, I can take care of adding InferenceClient snippets for many tasks in Python (similar to #971 but with your new structure).

Note: this change will break the script that generates the Inference API docs (here). I'll open a PR once this is merged/deployed.

[julien-c]

are the selected tab + sub-tab sticky? (not sure if they should be, cc @gary149)

[gary149]

are the selected tab + sub-tab sticky? (not sure if they should be, cc @gary149)

Currently no (I think it's fine)

[gary149]

Nice! I didn't try everything but I see an issue with some snippets like

http://localhost:5564/black-forest-labs/FLUX.1-dev?inference_api=true

[mishig25]

@gary149 8399207 should fix the issues you've mentioned

if you are testing on moon, make sure to build hfjs again by

cd packages/tasks
pnpm i
pnpm build