Update job information only when the transaction in ChunkOrientedStep is successful

[xeounxzxu]

Problem

• ChunkOrientedStep updated ItemStream and ExecutionContext inside the finally blocks of processChunkSequentially/processChunkConcurrently. Those blocks run even when a chunk rolls back, so the step persisted a “failed” offset. On restart, processing resumed after the failed chunk, silently skipping the data that had never been committed.
• ChunkOrientedStep updates ExecutionContext even when a chunk fails, leading to data loss on restart #5182

Solution

• Move the compositeItemStream.update and JobRepository.updateExecutionContext calls out of the chunk-processing finally blocks and into doExecute, immediately after each successful transaction finishes. Now the state is persisted only when a chunk commits; when a chunk fails and the transaction rolls back, the execution
  context remains unchanged, preventing data loss on restart.

[xeounxzxu]

@fmbenhassine Can you check the PR?