When an autopkgtest worker task is retried (and now passes), retrying the top-level regression analysis still reports a regression. The per-architecture regression analysis callback on the autopkgtest workflow is never re-triggered, so the autopkgtest workflow's output_data remains stale. The top-level QA regression analysis callback reads this stale data. ### Steps to reproduce - Upload a package that triggers an autopkgtest failure for a reverse dependency (e.g., goobox/amd64). - The regression analysis correctly flags this as a regression. = Retry the failing autopkgtest worker task - it now passes. - Retry the top-level "Regression analysis" on the QA workflow. - Expected: Regression analysis picks up the new passing result and no longer flags a regression. - Actual: Regression analysis still reports a regression. ### Example - Work request: https://debusine.debian.net/debian/developers/work-request/536185/ - goobox/amd64 was failing and correctly caught by regression analysis. - After retrying the autopkgtest, goobox/amd64 passed. - After retrying the regression analysis, it still reports a regression for goobox/amd64.