cmd/prometheus: add --auto-gomemlimit.refresh-interval flag

[sanidhyasin]

Which issue(s) does the PR fix:

Fixes #18712

Release notes for end users (ALL commits must be considered).

[FEATURE] Prometheus: Add `--auto-gomemlimit.refresh-interval` flag to periodically re-detect the container or system memory limit and update `GOMEMLIMIT` at runtime.

---

What this does

The auto-GOMEMLIMIT feature (--auto-gomemlimit) detects the cgroup or system memory limit only once at startup, via a single call to memlimit.SetGoMemLimitWithOpts. If the limit changes while Prometheus is running — for example, a Vertical Pod Autoscaler (VPA) adjusting a container's memory — GOMEMLIMIT keeps the value computed at startup until the process restarts.

This PR adds an optional --auto-gomemlimit.refresh-interval flag that wires the automemlimit library's existing WithRefreshInterval option. When set to a non-zero duration, the limit is periodically re-detected and GOMEMLIMIT is reapplied if it changed.

• Default is 0s, which preserves the current behaviour exactly (detect once at startup, no background refresh — WithRefreshInterval(0) is a no-op in the library).
• The flag is only used when --auto-gomemlimit is enabled, consistent with --auto-gomemlimit.ratio.
• Negative durations are rejected at flag-parse time by model.Duration, so the library's ticker can never receive an invalid interval.

Notes

• As discussed in the issue, the polling cost is negligible at typical intervals (≥30–60s) — it reads a single value from cgroup memory. The flag defaults to disabled so there is no behaviour change unless opted in.
• The help text calls out that a downward change in the limit (e.g. a VPA scale-down) can cause a temporary increase in GC activity, so operators aren't surprised.

Changes

• cmd/prometheus/main.go: new flag + config field, passed through as memlimit.WithRefreshInterval.
• docs/command-line/prometheus.md: regenerated via make cli-documentation.

I have signed the DCO.