fix(weave): preserve user IDs in object creation

[gtarpenning]

Summary

• preserve the authenticated wb_user_id across op, dataset, scorer, and evaluation creation
• propagate the ID to scorer and evaluation child ops
• convert the external user ID at the adapter boundary before ClickHouse persistence
• keep the in-memory trace server aligned with the production backend

Before / after

Before, a DatasetCreateReq carrying wb_user_id reached the backend helper, but the generated ObjSchemaForInsert hardcoded the field to None. The same happened for ops, scorers, and evaluations, while generated scorer/evaluation ops omitted the field entirely.

After, the adapter converts the external ID to its internal representation once and every object version emitted by these four creation paths stores that ID in both ClickHouse and the in-memory test backend.

Testing

• focused object-creation test against ClickHouse (passed)
• focused object-creation test plus the four existing no-ID create tests with --trace-server=fake (5 passed)
• uvx ruff check on all changed files
• uvx ruff format --check on all changed files
• pre-push hooks: ty, import-linter, pyright, mypy, Fixit, Ruff

Scope

This intentionally does not add global non-null assertions, change worker payload contracts, or backfill historical rows.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!