Best-effort teardown: drop the snapshot table, then delete the metadata FK-safe (join → MR → query → category).
Each delete is guarded individually — MR3 may have already deleted the MaterializedResult (and, via the server
subclass, its join row), so a delete that finds nothing must never skip the remaining cleanup (else the Query +
Category would leak on every run). IsSaved skips the row MR3 already removed; the per-delete catch handles the
join row the subclass deleted out from under the in-memory object.
Best-effort teardown: drop the snapshot table, then delete the metadata FK-safe (join → MR → query → category). Each delete is guarded individually — MR3 may have already deleted the MaterializedResult (and, via the server subclass, its join row), so a delete that finds nothing must never skip the remaining cleanup (else the Query + Category would leak on every run).
IsSavedskips the row MR3 already removed; the per-delete catch handles the join row the subclass deleted out from under the in-memory object.