fix: lock rows and skip them during concurrent poll

[thevaibhav-dixit]

Fixes #51

[Diggsey]

Thanks for the PR! I think this kind of change should be accompanied by a regression test though.

[bodymindarts]

Not sure how best to go about reproducing this locally. In the production environment we observed this we spawned ~200 jobs and had 2 workers running to get through the queue. Before this fix we consistently saw duplicate job_ids... now none anymore.

Do you have a tip on how you would go about testing this?

[Diggsey]

I would probably spawn a large number of jobs, and then start N threads, and have them all wait on a Barrier to make sure they are properly synchronized, before having them all simultaneously start the job runner.
If the same job is picked up twice, panic.
Rinse and repeat M times, where M is large enough to reliably reproduce the issue.