fix: correct cbgo_dcp_latency_ms_current metric type from Counter to Gauge

[nakrenat]

What

cbgo_dcp_latency_ms_current was exported as prometheus.CounterValue in
metric/collector.go, but its underlying value is a point-in-time latency
snapshot that is overwritten on every DCP event — not a monotonically
increasing total.

Why It Matters

Prometheus counters must never decrease. Because latency fluctuates
naturally (fast events produce small values, slow ones produce large values),
this metric regularly decreases between scrapes. This causes:

• Prometheus to emit stale-marker errors or silently drop decreasing counter samples
• rate() and increase() PromQL functions to return incorrect or zero results
• Grafana panels and alerts based on this metric to show misleading data

The inconsistency is easy to confirm by comparing the two latency metrics
side by side in collector.go:

// processLatency — correctly exported as GaugeValue
ch <- prometheus.MustNewConstMetric(s.processLatency, prometheus.GaugeValue, ...)

// dcpLatency — incorrectly exported as CounterValue  ← this PR fixes it
ch <- prometheus.MustNewConstMetric(s.dcpLatency, prometheus.CounterValue, ...)

[mhmtszr]

LGTM! Thanks for the pr.