feat(observability): collect process system metrics

[WenYuLuo]

What changed

• add an OpenTelemetry hostmetrics/process pipeline for long-lived AKernel and YuanRong processes
• collect CPU time/utilization, RSS/virtual memory, memory utilization, open file descriptors, thread count, and uptime every 15 seconds
• attach environment, role, pod, namespace, node, process, and normalized component resource attributes before Prometheus remote write
• normalize the truncated DataSystem process name to component_name="ds_worker"
• distinguish goruntime as frontend or scheduler from its stable runtime ID, then remove process.command_line before export
• pass pod and node identity into the collector for master, frontend, node, and standalone deployments
• provision an AKernel Process Resources Grafana dashboard with environment, node, Pod, role, and component filters

Dashboard

The dashboard provides six panels:

• CPU utilization by component
• resident memory by component
• open file descriptors by component
• thread count by component
• process uptime
• current resident-memory snapshot

Why

The collector previously accepted metrics pushed by applications but did not report operating-system resource usage for individual platform components. This left CPU, memory, file descriptor, thread, and uptime visibility unavailable at the component and node level.

The process allowlist intentionally excludes per-sandbox runsc processes to avoid unbounded time-series cardinality.

Validation

• validated both collector configurations with otel/opentelemetry-collector-contrib:0.120.0
• verified remote write end to end with shared-PID-namespace frontend, scheduler, and DataSystem test processes; Prometheus received frontend, scheduler, and ds_worker, with no command-line label exported
• parsed every dashboard JSON and all six new PromQL expressions with Prometheus 3.5.0 promtool
• provisioned the dashboard into Grafana 11.5.2 and verified its UID, title, and six panels through the Grafana API
• make SHELL=/bin/bash deploy-script-check
• Helm 3.18.4 lint and template rendering for the core and monitor charts
• git diff --check