Improve Zoekt reallocation logic
Background
Currently, Zoekt nodes' disk usage is a bit spiky, which triggers incidents. For example, 2024-04-13: kube_persistent_volume_claim_disk_s... (gitlab-com/gl-infra/production#17829 - closed)
Proposal
We need to investigate the cause and improve the logic so that