Need functionality to avoid lost update in jobs

Product: MDMCE 11.5.0 - Scheduler - Import jobs
See: Case TS001386519

The client is reporting "lost update" situations when 2 parallel jobs are processing updates to the catalog. This can happen when 2 jobs read an item before any of them has done a save (which would acquire a lock on the item). When one of the jobs saves the item and terminates (which releases the lock) the other job can then also save the item and overwrite any of the changes the first job has done (since only the changes on the initially read item are persisted). The risk of such "lost update" situations is even bigger with a settig of release_locks_early=true since then the locks are already released immediately after saving the item.
Example:
- job A reads item I1
- job B reads item I1
- job A changes item to I11
- job B changes item to I12
- job A persists item I11 and terminates
- job B persists item I12, overwriting I11.

This could be avoided by providing functionality to actively control locking, exactly how it is offered by a "reserve" functionality for users working concurrently:
- API for reserving an item by its primary key for the current job
(could return "false" if the item is locked by another process)
- API for releasing a lock for an item by its primary key for the current job

(Another option could be an "optimistic" schema: Every item object could hold a hidden timestamp which is compared at save-time to the currently persisted item. If the timestamps don't match then a concurrent update has been done in the meantime, making the memory item object stale. In this case the persisting would throw an exception and avoid overwriting an update).