Move shard_imports into memory-bound shard
Overview
We have a dedicated shard for the import jobs shard_imports this only runs a few jobs as seen below, running around 1 job per minute.
The imports shard has a custom configuration to allow long running jobs to finish without being interrupted and retried through a deployment: https://gitlab.com/gitlab-com/gl-infra/k8s-workloads/gitlab-com/blob/63aa392c38c7f3a332addfc01947e9af9236df41/releases/gitlab/values/gprd.yaml.gotmpl#L564-572
The memory-bound shard has similar characteristics and also very low traffic. We should make the memory bound shard support these long running jobs (by copying that configuration) and move the imports workload there to run more efficiently and have less noisy alerts on both low-traffic shards. This adjustment will also support long running export jobs on this shard better. To make sure we accommodate for new incoming traffic, we should also increase the maxReplicas for the memory-bound shard.
Once there are no more jobs getting scheduled on the imports queue, and everything is routed to the memory bound queue, we can remove the imports shard.