Update Transform MCP tool names and document the extraction tools

[simoncoombes]

The server renamed three of its tools, so the guidance we ship as extension context named tools that no longer exist. An agent following it called transform_files and got an unknown-tool error.

• transform_files is now start_transform_job
• check_transform_status is now check_job_status
• get_transform_results is now get_job_results

request_file_upload_url is unchanged. Names confirmed against the live server (Unstructured Transform 0.7.2) via its get_instructions tool, and they match the rename that already landed in the docs (Unstructured-IO/docs#1004).

The same release added structured data extraction, so a second commit documents that too rather than leaving the guidance describing only half of what the server does. The new section covers the order that matters: extraction reads the element JSON a parse produces, so parse first and carry the output_ref into suggest_extraction_schema_for_file (when the user has no schema) and then start_extraction_job. It also spells out the two things agents get wrong otherwise, that results come back inline rather than behind a download URL, and that they carry provenance which should be kept rather than unwrapped to bare extracted_data.

I ran the flow against production before writing it up (parse an image with strategy=vlm, draft a schema from the resulting output_ref, extract, read back provenance-wrapped JSON), so the documented sequence matches real behaviour.

scripts/validate.sh passes. This touches a different part of AGENTS.md than #2, so the two do not conflict.

[cubic-dev-ai]

No issues found across 1 file

_{Re-trigger cubic}