[InferenceSnippet] do not fail entirely if snippets did not work

[Wauplin]

Currently doc-builder fails on Inference Provider docs with:

(taken from PR huggingface/hub-docs#1788 / see CI here)

Failed to get provider helper for replicate (automatic-speech-recognition) InferenceClientInputError: Task 'automatic-speech-recognition' not supported for provider 'replicate'. Available tasks: text-to-image,text-to-speech,text-to-video

This is due to provider Replicate having a automatic-speech-recognition model listed even though the JS / Python clients do not support it (and therefore we cannot generate the snippets). Instead of failing entirely it would be best to warn on error and continue. This allows other snippets to be correctly updated.

---

Note: a solution in parallel is to unlist the failing model on Replicate side. I already pinged them on Slack about that but in any case, we should not block doc updates just because of a mistake from a provider.

[Copilot]

Pull Request Overview

This PR ensures that errors during inference snippet generation do not block documentation updates by catching exceptions and returning empty snippets. Key changes include:

• Adding a new interface (InferenceProviderMappingEntry) to support mapping information.
• Creating a helper function (getInferenceSnippetsOrEmpty) that wraps snippet retrieval in a try-catch.
• Updating code calls to use the new helper function, preventing complete failure on snippet generation failures.

Comments suppressed due to low confidence (1)

kit/src/lib/InferenceSnippet/InferenceSnippet.svelte:76

• Add a brief comment explaining the purpose of getInferenceSnippetsOrEmpty(), particularly why errors are caught and an empty array is returned.

	function getInferenceSnippetsOrEmpty(

[Wauplin]

@mishig25 PR is ready for review