ci: classify current issues and pull requests with Qwen

[Xuanwo]

Which issue does this PR close?

No linked issue. Replaces the retired hosted labeling automation.

Rationale for this change

PRs need release-note categories, and issues need type and component labels. Classify the current item with Cloudflare Workers AI using the Qwen model tested against OpenDAL examples. Keep the component catalog aligned with the repository so valid answers such as services/goosefs are available.

What changes are included in this PR?

• Add event-driven labeling for opened, edited and reopened issues/PRs, plus PR synchronization. Process only the event's open item; no scheduled scan or backfill.
• Separate issue types from PR release-note categories, validate component labels, detect explicitly declared breaking changes, and calculate PR size from changed lines.
• Create missing label definitions from the service inventory and binding/integration directories. Preserve existing names and metadata.
• Preserve existing classifications and maintainer-controlled labels. Component/breaking labels are additive; size labels are replaced.
• Document label policy and setup, and add event/validation contract tests to Check CI.

The privileged PR workflow checks out only the default branch. PR code is never executed.

Are there any user-facing changes?

New or subsequently updated issues and PRs receive labels according to .github/LABELS.md. Existing history is not scanned.

Deployment requires the CLOUDFLARE_ACCOUNT_ID repository variable and CLOUDFLARE_API_TOKEN secret. The account variable and dedicated Workers AI Read API token secret are configured. End-to-end event-driven labeling in Actions remains to be verified after this workflow reaches the default branch. The final request format was exercised through the connected Cloudflare account on six real OpenDAL examples, and its returned responses passed the implementation's parser/validation. End-to-end Actions inference and label assignment with the workflow token remain unverified.

Classification remains approximate: testing found sensitivity to title prefixes and occasional component omissions. Maintainers can correct classifications; the workflow preserves existing primary labels.

Breaking changes

AI Usage Statement

• Harness: Codex desktop. Model: GPT-6; exact identifier and reasoning effort are not exposed in this session. Role: investigated the retired workflow, evaluated replacements, implemented the classifier/workflow/tests, and drafted this PR.
• Harness: Cloudflare Workers AI through its MCP API. Model: @cf/qwen/qwen3-30b-a3b-fp8. Effort: /no_think, temperature 0. Role: classified real PR/issue inputs during validation. It did not execute code or publish GitHub comments.