Improve orphan final artifacts finder rake task performance

This is adding some minor improvements to the rake task that generates the list of orphan objects.

This is based on feedback from https://gitlab.com/gitlab-com/gl-infra/production/-/issues/17579#note_1771422415:

I've reviewed the logic, and I suspect the bucket scan isn't in fact the bottleneck. We're most likely limited by the job_artifact_record_exists? call. Increasing concurrency will put too much pressure on postgres.

I think the main optimization we should look into is batching up those calls. Basically, check existence of 1k job artifact records in a single query.

I also added some minor changes to batch size and redis marker TTL that I forgot to add before. This was from the suggestions in gitlab-com/gl-infra/production#17383 (comment 1738630685).