Excavate worker version IDs from corpus exports

Listen to this noise generator.
Remove all the foreign keys from ML results to worker_version in arkindex/documents/export/structure.sql. Keep the worker_version table, and keep worker_run.worker_version_id. Those will still be required by load_export.
Update worker_version.sql so it uses worker_run_id from every ML result in the large UNION, and only then joins on process_workerversion.
Remove the worker version IDs from all the other export queries that used to provide this column.
Bump the export version, if that wasn't already done in this release.
Update the maximum supported version in load_export to the bumped export version, if necessary. The minimum remains at 8 since we still can ignore the deleted columns and tables.
Remove worker version import support from load_export.