Port v5-next OPFS/sqlite-worker backlog to next (re-implementation needed)

[PhilWindle]

The storage / OPFS (kv-store) slice of the v5-next → next backlog cannot be cherry-picked: v6 (next) re-architected the OPFS / sqlite-worker layer, so these are re-implementations against the new architecture, not merges. There are no clean commits, hence a tracking issue rather than a PR.

Each needs porting by the storage owner:

• fix: release OPFS handles before sqlite worker close/delete ack #24647 git cherry-pick -x af3323bf4a — fix: release OPFS handles before sqlite worker close/delete ack
• fix: handle corruption errors #24739 git cherry-pick -x 007dcc2ae6 — fix: handle corruption errors
• feat: weblock controlled opfs pool #24740 git cherry-pick -x 386f120fb2 — feat: weblock controlled opfs pool
• chore: handle legacy duplicate opaque handles #24743 git cherry-pick -x cfbd62af4c — chore: handle legacy duplicate opaque handles

Context: these were merged to v5-next after the ~2026-07-08 cut that reshaped next. Part of the manual v5-next → next backlog sweep.

[mverzilli]

To avoid future confusion: the "v6 (next) re-architected the OPFS / sqlite-worker layer, so these are re-implementations against the new architecture, not merges" statement is false. There's a missing PR in the series that causes this misinterpretation: #24631. Working on a port of everything at once.