In this ticket we aim to improve the acquire/release semantics of the serverless locks and the tenant access blockers. There are two main issues to address :

• The rollback handler for MTAB and serverless lock is only registered on primaries. This is a bug as there might be a write conflict exception on secondaries that would cause them to retry the write (and reacquire mtab/serverless lock).
• The rollback handler is registered some time after acquiring the serverless lock (https://github.com/10gen/mongo/blob/347fcaa2fc3b26441f622cddae8ea06553cb7786/src/mongo/db/repl/tenant_migration_donor_op_observer.cpp#L70). If there is an error thrown inbetween, we might retry the operation without releasing the lock.

The fixes :

• Remove `if (onPrimary)` conditions.
• Move `rollback` handler as close to acquire serverless lock and mtab (and decouple the handlers).
• Review all places we use `acquireLock` to ensure correct behavior.