fix(wa-sqlite): recover cleanly from OPFS initialization failure

[marcus-pousette]

Problem

#210 prevents the reproduced automatic-extension failure, but any remaining OPFS initialization error still has to unwind every layer cleanly. A failed attempt can leave behind a database handle, VFS resources, wa-sqlite module callbacks/retry state, worker listeners, pending RPC calls, or an attached SharedWorker port.

In SharedWorker mode, that stale port can keep a failed session alive and make the next client using the same worker name fail as if a different database were already initialized.

Change

• Close partially initialized database and VFS resources idempotently.
• Run optional memory fallback in a fresh wa-sqlite module. Closing the failed VFS does not unregister its callbacks or reset module retry state, so reusing that module is unsafe.
• Preserve both the OPFS error and the fallback error when both attempts fail.
• Remove listeners, reject pending RPC calls, close ports, and terminate dedicated workers when initialization, close, or drop fails.
• Cover recovery by forcing an OPFS open failure and then successfully reusing the same SharedWorker name.

This PR is the cleanup and recovery layer after #210; it does not claim to be the root fix for the initialization failure.

Dependencies and merge order

Stacked directly on #210. Use a merge commit after #210, then retarget #163 to main.