Madvise settings rollout

See #318 (moved)

We are experimenting with reverting the madvise hint to MADV_DONTNEED which should report RSS figures in a way closer to the memory that is actually consumed, which we think will help with pod memory limits not being reached as quickly.

We have enabled this for gstg and cny here: gitlab-com/gl-infra/k8s-workloads/gitlab-com!672 (merged)

After reviewing this change once live, we should either:

• Make the same change to gprd (successful outcome)
• Revert these changes (unsuccessful outcome)

Relevant dashboards:

• https://dashboards.gitlab.net/d/websockets-main/websockets-overview?viewPanel=172578411&orgId=1
• RSS vs Go heap stats in Thanos