fix(metrics): count nostr_events_total on successful DB write, not on ingress

[archief2910]

Description

nostr_events_total{kind="…"} was incremented in ingesterProcessEvent immediately after parse/verify, before duplicate detection, write-policy plugin, and the writer pipeline. That counted every accepted parse toward kind totals, including events that never became EventWriteStatus::Written (duplicates, plugin rejects, replace/delete outcomes, etc.).

This PR:

• Removes PROM_INC_EVENT_KIND from src/apps/relay/RelayIngester.cpp.
• Adds PROM_INC_EVENT_KIND(std::to_string(packed.kind())) in src/apps/relay/RelayWriter.cpp only when newEvent.status == EventWriteStatus::Written, alongside writtenEventsTotal.inc().
• Updates the Prometheus # HELP nostr_events_total text in src/PrometheusMetrics.h to state that the counter reflects events persisted to the DB, not raw client ingress.

Files: src/apps/relay/RelayIngester.cpp, src/apps/relay/RelayWriter.cpp, src/PrometheusMetrics.h

Motivation and context

Operators using Prometheus to reason about stored traffic and kind mix should not inflate totals with duplicates and write-path rejects. Counting at Written matches that intent.

Types of changes

• Non-functional change
• Bug fix
• Breaking change
• Breaking change

Checklist

• HELP text updated
• Build passes locally

[hoytech]

Technically a breaking change, but I doubt anyone is relying on this behaviour. And I agree it's more useful/accurate this way.