feat: Metrics endpoint

[va4st]

Summary

Add embedded Prometheus monitoring to mtproto-proxy and provide a ready-to-use Docker monitoring stack with Prometheus + Grafana.

related #169

This PR adds:

• version module. all version consts moved into separated constant module to maintain it via one parameter
• an in-process /metrics endpoint on a dedicated configurable port
• global proxy traffic counters
• optional cgroup RAM metrics for container environments
• per-user traffic and active connection metrics
• Grafana dashboard JSON for import
• Docker Compose example and monitoring docs

What Changed

Embedded Prometheus endpoint

• Add [monitoring] config section:
  • enabled
  • host
  • port
• Start a dedicated HTTP listener inside mtproto-proxy
• Expose GET /metrics in Prometheus text format

Global proxy metrics

• Active connections
• Handshakes in flight
• Accepted / closed totals
• Drop counters
• Pause state gauges
• Global traffic counters:
  • mtproto_client_to_upstream_bytes_total
  • mtproto_upstream_to_client_bytes_total

Memory metrics

• Keep process metrics:
  • process_resident_memory_bytes
  • process_virtual_memory_bytes
  • process_cpu_seconds_total
  • process_open_fds
  • process_max_fds
• Add optional cgroup memory metrics when available:
  • mtproto_cgroup_memory_usage_bytes
  • mtproto_cgroup_memory_limit_bytes

This works both on bare metal and in containers:

• bare metal: process_* metrics still work
• containers: cgroup memory metrics appear when exposed by the runtime

Per-user metrics

Pre-create per-user metric records from configured users and expose:

• mtproto_user_connections_active{user="..."}
• mtproto_user_client_to_upstream_bytes_total{user="..."}
• mtproto_user_upstream_to_client_bytes_total{user="..."}

Per-user byte counters are updated on the actual write path, so they match the semantics of global traffic totals.

Dashboard and docs

• Add Grafana dashboard JSON with:
  • memory usage
  • total throughput
  • total transferred
  • connection stats
  • traffic throughput
  • per-user active connections
  • top users by throughput
  • top users by transferred volume
• Add hack/docker example with:
  • docker-compose.yml
  • prometheus.yml
  • dashboard JSON
  • README walkthrough
• Link the Docker monitoring guide from the main README

Why

This makes monitoring usable in the common Docker deployment model without relying on the existing Python dashboard stack.

Key goals:

• one proxy process exports its own metrics
• no log parsing for metrics
• no extra exporter sidecar required
• works in both Docker and raw system deployments
• supports operator visibility by configured user

Testing

Verified with:

• zig build test
• zig build
• dashboard JSON validation via jq empty
• compose file validation via docker compose config
• deployment on staging with real load and usage

Notes

• Counter totals are process-lifetime counters and reset on proxy restart, which is expected for Prometheus counters.
• cgroup memory metrics are best-effort and only appear when the relevant cgroup files are available.
• Per-user metrics are based on users from the current loaded config and update after proxy restart when config changes.

[va4st]

@sleep3r Нужно ревью, пожалуйста)