Allow scheduling new job manually

[jpmckinney]

The old instructions to delete jobs in order to allow new jobs to run can cause all sorts of issues. Namely, deleting a job could delete a Kingfisher Process collection or Pelican dataset while work is still queued in RabbitMQ, which could cause 100,000s of errors to be reported to Sentry.

We can instead have a process where a job can be manually scheduled, and delete those instructions. There are two scenarios to consider:

• Last job is complete, but we want to run a new job sooner than scheduled.
  • We can check that the collection isn't already is_out_of_date (to avoid duplication by manageprocess if e.g. an admin unfreezes a publication, and is then in a rush to schedule it, instead of waiting 5 mins), and then create the job. (This scenario also checks that no job is running.)
• Last job is running.
  • We can require admins to cancel the job Add cancel action to job admin #352, and then follow the first scenario.

Update the docs that refer to this issue (#350) and surrounding text.

Add relevant tests first #128

Blocked by #352