Allow worker to specify which arguments to include in uniquing hash

[sax]

We're running into an issue where we want to add non-unique arguments to a sidekiq worker, but we'd like the job to still be counted as unique. For instance (pseudocode):

MyWorker.enqueue_delayed(product.id, Time.now, 1.hour.from_now)
sleep 10
MyWorker.enqueue_delayed(product.id, Time.now, 1.hour.from_now)

Time.now is obviously not going to be unique, and so a second job will be scheduled. What I'd really like is to be able to specify which arguments should be included in the hashing algorithm, so that the second job would not be enqueued.

I'm happy to implement this, but wanted to ensure that it was something that would be accepted as a pull request before doing so.

[sax]

We started using this in production yesterday, with the use case shown above. We run a lot of sidekiq jobs that are either computationally expensive or expensive on our databases. For these types of jobs, we like to schedule them some amount of time in the future, with payload uniquing.

The problem is that sometimes we need to include data in the job payload that makes the job unique even when we'd like it to fold into a previously queued job. For instance, when queuing a change to a user feed, the timestamp of the original job may be needed in order to correctly do work (by doing work on the delta between timestamp and now). New changes to the same user feed queued before the first job runs may be identical to the first job, except for the timestamp. Only the first job needs to actually run in order to correctly update the feed.

This can also help to keep queue counts from exploding when workers need to be stopped for some reason, or when due to various issues queues back up.