feat(huggingface): Add Hugging Face AI provider

[filipbech]

Summary

• New Umbraco.AI.HuggingFace provider package — chat-only, talks to the Hugging Face Inference Providers router (https://router.huggingface.co/v1).
• Reuses Microsoft.Extensions.AI.OpenAI (no new SDK dependency, no dependency on Umbraco.AI.OpenAI — both packages just consume the same NuGet).
• Validates credentials against the Hub's whoami-v2 endpoint because the router's /v1/models is fully public and would otherwise mask invalid tokens in Test Connection.
• Registers the package in Umbraco.AI.slnx, both install scripts (sh + ps1), the package-test-site scripts, and the level1Products matrix in azure-pipelines.yml.

Out of scope (future)

• Embeddings — the router's OpenAI-compatible endpoint is chat-only. Adding embeddings would require either the native HF Inference API or the community tryAGI/HuggingFace SDK. Noted in the package's CLAUDE.md.
• Image gen, speech, classification — no AI*CapabilityBase exists for these in Core.

Test plan

• dotnet build Umbraco.AI.HuggingFace/Umbraco.AI.HuggingFace.slnx — green, lock file generated, packages.lock.json confirms no Umbraco.AI.OpenAI dependency
• Demo site smoke test — provider appears in the Connections dropdown, model list populates from GET /v1/models, chat round-trip works with a real hf_... token
• Test Connection — fails with invalid/missing token (whoami-v2 probe), passes with valid token
• Reviewer to verify the install scripts on Windows (only sh tested locally)

🤖 Generated with Claude Code

[filipbech]

chat only