Harden AWS transaction locking mechanism to support long running I/O operations

[homar]

Currently, AWS transaction log locking assumes that operation under lock will return within 5 minutes.

After 5 minutes lock is assumed expired and if file was written after that grace period we may have conflicting writes from two transactions.

The mechanism can be improved by refreshing the lock in separate thread, while operation under log is in progress.

[wendigo]

Closing per #23145