feat: sync HTTP types and auth with Node SDK 2.2.0

[suguanYang]

Summary

• Align the Python SDK with Node 2.2.0 HTTP/type parity: document_metadata on job create/parse (official created_by_client=python-sdk / client_version defaults; caller keys win), planned document_id on Job create responses, and document_metadata on Document.
• Add documents.get_page_citation_source and typed PageCitationAsset for chunk.metadata["pageAssets"]. The SDK does not generate citation assets.
• Accept auth_token_provider for short-lived bearer tokens. A static api_key still wins; the provider is called per request.

Closes #34 for the P0 HTTP/type slice. Knowledge/parsed-storage (P1) is out of scope. Do not merge release PR #32 first — release-please should rewrite it so 2.0.0 includes these commits.

Test plan

• uv run ruff check src/ tests/
• uv run mypy src/knowhere
• uv run pytest -q
• Create a job with no document_metadata and confirm the body includes created_by_client=python-sdk and client_version
• Override created_by_client and confirm the caller value wins
• Confirm Job.document_id is populated when the API returns a planned id
• Call documents.get_page_citation_source and confirm GET /v2/documents/{id}/files/page-citation-source
• Construct Knowhere(auth_token_provider=...) without api_key and confirm the Authorization header refreshes per request

Made with Cursor