feat(ratelimit): track per-window token usage and utilization

Poll /api/oauth/usage every 5 min and extract utilization from
/v1/messages response headers for real-time updates. Track proxy
tokens in/out per rate limit window (5h/7d), resetting on window
change. Expose as OTel observable gauges for Grafana dashboards.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

main.go

@@ -14,6 +14,7 @@ import (
 	"github.com/fujin/anthropic-proxy/internal/config"
 	"github.com/fujin/anthropic-proxy/internal/logging"
 	"github.com/fujin/anthropic-proxy/internal/proxy"
+	"github.com/fujin/anthropic-proxy/internal/ratelimit"
 	"github.com/fujin/anthropic-proxy/internal/server"
 	"github.com/fujin/anthropic-proxy/internal/telemetry"
 	"github.com/rs/zerolog/log"
@@ -25,8 +26,17 @@ func run() error {
 		return fmt.Errorf("load config: %w", err)
 	}
 
+	// Create usage tracker (started later once credential is loaded)
+	var credForTracker *auth.Credential
+	tracker := ratelimit.NewTracker(func() string {
+		if credForTracker == nil {
+			return ""
+		}
+		return credForTracker.Token()
+	})
+
 	// Initialize telemetry (metrics always active; OTLP export when endpoint set)
-	telemetryShutdown, logBridge, err := telemetry.Setup(context.Background(), cfg.Telemetry)
+	telemetryShutdown, logBridge, err := telemetry.Setup(context.Background(), cfg.Telemetry, tracker)
 	if err != nil {
 		return fmt.Errorf("telemetry setup: %w", err)
 	}
@@ -85,6 +95,8 @@ func run() error {
 
 	log.Info().Str("credential", cred.Email).Msg("credential loaded")
 
+	credForTracker = cred
+
 	pool := auth.NewPool([]*auth.Credential{cred})
 
 	ctx, cancel := context.WithCancel(context.Background())
@@ -92,6 +104,7 @@ func run() error {
 
 	pool.RefreshExpiring(context.Background())
 	auth.StartBackgroundRefresh(ctx, pool)
+	tracker.Start(ctx)
 
 	var profile *proxy.SniffedProfile
 	if cfg.ClaudeBinary != "" {
@@ -103,7 +116,7 @@ func run() error {
 	}
 
 	log.Info().Int("port", cfg.Port).Msg("starting server")
-	srv := server.New(cfg, pool, profile)
+	srv := server.New(cfg, pool, profile, tracker)
 
 	quit := make(chan os.Signal, 1)
 	signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)