v2.9

Major release: uvlhub is rebuilt on splent_framework and gains dataset versioning - replace a UVL and publish a new, linked Zenodo version.

Added

• Dataset UVL versioning. Replace the UVL of a draft in place, or publish a new version of an already-published dataset. A new version creates a fresh Zenodo version (its own DOI, linked to the previous one through the shared concept DOI) and a new local dataset linked to the previous via dataset_origin_id.
• Version history on the dataset page and a version badge in "My datasets"; the list now shows only the latest version of each lineage.
• Dataset edit mode now manages UVL files (replace draft files; publish a new version for published datasets).

Changed

• Rebuilt on splent_framework. app/modules/ -> app/features/, the bespoke core is removed, dependencies are pinned in pyproject.toml, and the app runs from /workspace. The rosemary CLI, GitHub Actions and docs were aligned to the new layout.
• Every feature refactored to SOLID services/repositories with thin routes and the splent testing pyramid (unit / repository / service / integration / e2e / load); the hub <-> UVL-domain coupling is inverted via splent signals.
• Editing a synchronized dataset without an explicit type no longer clears its Zenodo DOI.

Fixed

• Production Docker build/deploy. The webhook service no longer creates a Docker client at import time (which broke docker build via rosemary webpack:compile); rosemary's selenium/locust commands use the /workspace layout.
• Generator wizard now boots its self-hosted Pyodide runtime and the dataset description editor (TinyMCE) loads its assets - both were 404ing on nested asset paths.

Upgrade notes

• Data-preserving Alembic migrations are included and apply automatically on deploy (flask db upgrade, run by the production entrypoint).
• Ensure WORKING_DIR=/workspace in the production .env.

Full changelog: v2.8...v2.9