Refacto Inference snippets tests (autogeneration + 1 snippet == 1 file)

[Wauplin]

This PR is a refacto of the inference snippet tests. The main goal is to reduce friction when updating the inference snippets logic. In particular:

• each generated snippet is saved into a single test file with the correct file extension => should greatly improve reviews. Currently snippets are saved as hardcoded strings in JS files, making them hard to review (no syntax highlighting, weird indentation).
• snippets can be updated with a script, reducing the need to manually edit tests (which is quite time-consuming)

This PR is quite large given the auto-generated files. The main parts to review are:

• generate-snippets-fixtures.ts => the script that generates the snippets and test them.
• package.json, pnpm lock files, etc. => I'm not entirely sure of what I did there (especially to make tasks-gen depend on tasks)
• python.spec.ts / js.specs.ts and curl.specs.ts have been removed
• everything in packages/tasks-gen/snippets-fixtures/ => the test cases. I've only committed the ones that where previously tested.

Thanks to this PR, I fixed a typo in the JS snippets (a missing comma) and curl snippets (consistency between " and ').

cc @mishig25 with whom I quickly discussed this

[mishig25]

for the info, this is how the same thing is declared in moon. In subseq PR, we can probably declare InferenceSnippetLanguage in tasksand use it in tasks-gen & moon

export const inferenceSnippetLanguages = ["python", "js", "curl"] as const;

export type InferenceSnippetLanguage = (typeof inferenceSnippetLanguages)[number];

[mishig25]

lgtm !
thanks for this refactor

[Wauplin]

thanks for the review :)