[Bot] Update Inference Providers documentation

[HuggingFaceInfra]

This PR automatically upgrades the @huggingface/tasks and @huggingface/inference packages and regenerates the Inference Providers documentation by running:

cd scripts/inference-providers
pnpm update @huggingface/tasks@latest @huggingface/inference@latest
pnpm run generate

This PR was automatically created (and merged) by the Update Inference Providers Documentation workflow.

---

Note

Low Risk
Generated documentation and dependency pin only; no runtime or API implementation changes in this repo.

Overview
Bumps @huggingface/tasks from 0.21.25 to 0.21.28 in scripts/inference-providers and regenerates Inference Providers markdown so InferenceSnippet examples and recommended-model lists match current warm provider mappings.

Chat / VLM: VLM doc examples move off MiniMaxAI/MiniMax-M3 to provider-specific models (e.g. thinkingmachines/Inkling, moonshotai/Kimi-K2.7-Code, bottlecapai/ThinkingCap-Qwen3.6-27B). Groq drops its VLM snippet and switches the LLM example to openai/gpt-oss-20b; ovhcloud aligns on the same smaller OSS model. PublicAI changes the LLM example to Bielik and adds a VLM snippet for Apertus v1.5.

HF Inference task docs: Example models change for image classification, segmentation, and zero-shot classification; text-classification gains a recommended language-detection model. Chat-completion / image-text-to-text recommended VLM lists add Qwen/Qwen2.5-VL-3B-Instruct.

^{Reviewed by Cursor Bugbot for commit 680cda5. Bugbot is set up for automated code reviews on this repo. Configure here.}

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.