[ARI-30] Graceful error propagation across the pipeline

[bjridicodes]

Linear: ARI-30

Type: ⚙️ Enabler | Sprint: 6 | Points: 3

As a Data Platform OPS engineer, I want pipeline errors to be caught, logged, and recorded in PipelineState so that a failure in one agent does not crash the entire process or leave the state unrecoverable.

Acceptance Criteria

Given Agent 2 raises LogStoreUnavailableError
When the pipeline handles the error
Then PipelineState.error is set with the exception message
And Agent 3 receives an empty log result (not an exception)
And a notification is still sent informing the engineer that no logs were available

Given an unhandled exception occurs in any agent
When the pipeline catches it
Then it is logged as ERROR with full traceback
And the incident number and agent name are included in the log entry

Given the pipeline processes 10 consecutive incidents with no errors
When logs are reviewed
Then no WARNING or ERROR entries appear (clean run validation)