fix: report job count only after it was persisted

[bajtos]

Fixes CheckerNetwork/node#583 (hopefully).

This change has a side effect when running a high-job-frequency module like the former Saturn L2 node or Voyager:

• Before, we would update the state file every time a job was completed.

• Now, we update the state file at most once per 200ms, as configured via job_report_delay:

  https://github.com/filecoin-station/zinnia/blob/9d882e48e5b9f0fdab2530fdc08c69987a4237a6/daemon/main.rs#L78-L82

I am arguing this is a change for the better because there is no point in hammering the file system (and SSD drives) with too frequent updates of the state file.

[bajtos]

@juliangruber ping, do you have any opinion on this change? If not, then I propose to proceed to land & publish it.