Move scheduling and cancelation db operation to atomic update requests

[badtst]

Description

Use atomic UPDATE statements for job cancellation/job scheduling instead of WITH FOR UPDATE statements which are noop on sqlite.

Atomic update statements are preffered to other solutions like exlusive transactions as they have a reliable behavior across the different backends and the logic performed within them is actually quite simple.

Improved testing for those possible race conditions:

• introduce repeated tests where we concurrently run cancellation logic/scheduling logic
• tests where we purposely interweave a cancel request in the middle of the scheduling logic

Changed

• Minor refactor of the scheduler strategy logic for better separation of responsabilities
  • Each scheduling strategy is tasked with identifying the next schedulable job id
  • get_next_job interface tries to claim the job by setting scheduled_at datetime of the previously identified job id

[MatthieuMoreau0]

nit: Can we get rid of these chain of thoughts statement which are just remnant from debugging? Let's focus on what has been implemented and why