Re-point a v3 dashboard at a different dataset

Swap the data underneath an existing board. The board is its own template: its current canvas carries the field roles, and they are resolved onto the new dataset with the same deterministic machinery the apply path uses — no LLM. Send a mapping (the shape /v3/templates/resolve-mapping returns) to use confirmed pairings, or omit it and the server proposes one.

When the new dataset carries the SAME columns — a refreshed extract, the common case — the board keeps the columns its author chose rather than re-scoring and possibly picking differently.

The swap lands as an ordinary canvas version, so undo, revert and version-restore all cover it with nothing extra. Sources are REPLACED, never added, so a board never ends up reading two datasets. Asynchronous because it re-bakes.