feat: Grafana dashboard + richer per-proxy metrics

[almeidapaulopt]

Problem

TSDProxy exposes Prometheus metrics via MetricsHandler() (internal/proxymanager/proxymanager.go), but there is no pre-built Grafana dashboard. The existing metrics are sparse — HTTP request counts and latency aren't tracked per-proxy. UDP relay has per-source rate limiting and client tracking but none of that is exported as metrics. Operators have limited visibility into:

• Per-proxy request rates and latency
• Per-proxy error rates and bandwidth
• Active connections per proxy
• UDP relay client counts
• TLS certificate expiry
• Tailscale connection status
• Memory/goroutine pressure from proxy count

Proposed Solution

Part 1: Richer metrics export

Add per-proxy metrics to the existing internal/core/metrics package:

• tsdproxy_proxy_requests_total{proxy, port, status} — request count by proxy/port/status code
• tsdproxy_proxy_request_duration_seconds{proxy, port} — latency histogram
• tsdproxy_proxy_up{proxy} — 1 if healthy, 0 if down
• tsdproxy_proxy_connections_active{proxy, port} — active TCP connections
• tsdproxy_udp_clients_active{proxy, port} — concurrent UDP clients
• tsdproxy_proxy_status{proxy} — current status enum (running/paused/error/etc.)
• tsdproxy_cert_expiry_seconds{proxy} — TLS certificate remaining lifetime

Part 2: Pre-built Grafana dashboard JSON

Ship a grafana/dashboard.json in the repo that visualizes:

• Per-proxy health grid
• Request rate and latency heatmap
• Error rate by proxy
• Active connections overview
• UDP client counts per proxy
• Proxy lifecycle events
• Overall system status

Implementation Notes

• The Metrics struct at internal/core/metrics uses Prometheus client — adding new counters/histograms is straightforward
• The HTTP reverse proxy in internal/proxymanager/port.go already passes proxyName and portName to m.Middleware()
• Net/http transport-level instrumentation could use otelhttp (already wired when telemetry is enabled) for additional span-level metrics
• The dashboard JSON should be kept under version control in a grafana/ directory at repo root

Alternatives

• Prometheus + Grafana is already a common stack; no need for a custom metrics backend
• OpenTelemetry is already supported — metrics could also be exported via OTLP, but Prometheus is the most universally expected