feat(server): deduplicate library and metadata jobs

[etnoy]

This adds jobs ids to several jobs which reduces duplicated work. For example, a large library refresh might go on for such a long time that a cron job kicks in and starts refreshing it before the first refresh finishes. This PR prevents it from happening.

I've also added job ids to metadata extraction and thumbnail generation, so we for instance don't queue a thumb generation for the same asset more than once.

[danieldietzler]

Code LGTM

[mertalev]

Can you set a job ID only for the queue job and not for the asset-level ones? It should be essentially the same result but much lower impact on queueing behavior.

[etnoy]

Can you set a job ID only for the queue job and not for the asset-level ones? It should be essentially the same result but much lower impact on queueing behavior.

The actual work-doing jobs of thumb generation and metadata extraction are the ones that will spend the absolute majority of time in the job queue for large libraries. The queueing jobs dwindle in comparison, and having them only check so that a queueing job already exists makes no difference to what we already have today.

After thinking about this some more I'm more inclined to think that the performance penalty of individual job ids should be worth it. In my current round of testing I've imported 500k assets and after two days I still have 350k thumbs left to generate. That's on a powerful Xeon server with 32 cores. Had I not disabled cron library scanning it would have queued the thumbnail refresh several times over by now.

[mertalev]

I'm really not in favor of solving this by multiplying the number of requests by 1000x or even 10000x. Maybe you can check at the start of the queue job if there are any library jobs and abort? This doesn't necessarily need to be solved with job IDs.