[mnapoli]

Follows #214 (merged).

Problem

BrefServiceProvider registers a JobFailed listener that stores failed jobs with the failed job provider. It registers this listener in every process: the Lambda queue worker, but also the web functions, and php artisan queue:work / queue:listen / Horizon on a developer's machine.

php artisan queue:work already stores failed jobs itself (WorkCommand::logFailedJob(), also a JobFailed listener). So in local development, each failed job is stored twice. With Laravel's default database-uuids driver, the second insert fails on the unique uuid column, and that error replaces the job's own exception (see the previous PR for why). This is php artisan queue:work in a Laravel 13 application with this package installed:

local.INFO: Processing job 1 {"name":"App\\Jobs\\FailingJob"}
  App\Jobs\FailingJob .. RUNNING
local.INFO: FailingJob local: step 1
  App\Jobs\FailingJob .. 2.72ms FAIL
local.ERROR: SQLSTATE[23000]: Integrity constraint violation: 19 UNIQUE constraint failed: failed_jobs.uuid (Connection: sqlite, ..., SQL: insert into "failed_jobs" ...)

The job's exception is never logged. With the database driver (no unique uuid), the job is stored twice instead.

Two smaller issues come from the same listener:

• It is registered in enableDetailedJobLogging(), so with bref.log_jobs set to false, failed jobs are not stored at all on Lambda. That option is documented as "disable detailed logging of every job execution".
• It also stores failed sync jobs (SyncQueue calls Job::fail() too). A standard Laravel application does not store them: the exception is thrown to the code that dispatched the job.

Fix

QueueHandler, the Lambda equivalent of queue:work, now stores failed jobs, the same way WorkCommand does: it registers the JobFailed listener when it is created. The listener keeps the behavior of the previous PR: if the failed job cannot be stored, it reports that error and does not throw it.

Behavior changes:

• php artisan queue:work / queue:listen / Horizon: each failed job is stored once, by Laravel.
• Lambda: failed jobs are stored, whatever the value of bref.log_jobs.
• Failed sync jobs are no longer stored, like in a standard Laravel application. (On the queue worker function, a sync job dispatched from inside a queued job is still stored, as with queue:work.)
• The failed job provider is resolved only when a job fails, not on every boot of the application (including HTTP requests).

Tests

• QueueHandlerTest::testTheFailedJobIsStored: the Lambda handler stores the failed job with the right connection, queue URL and exception.
• QueueWorkCommandTest: php artisan queue:work with the database-uuids driver stores the failed job once, and only the job's exception is reported. This test fails before the change with the UNIQUE constraint failed: failed_jobs.uuid error.

Tested with Laravel 10, 11, 12 and 13, and on Lambda: a job with $tries = 1 that throws has the same logs as with the previous PR, and its SQS message is deleted.