feat: 余额预扣 reserve/capture、冷却兜底重试、Prometheus /metrics

- 预扣：每请求按 body 估算成本上界计入冻结额（1e-8 USD 原子计数），402 判定
  改为「余额 - 在途冻结 ≤ 0」，并发的长流式请求不再互相透支；OT_PROXY_RESERVE_MAX_TOKENS
  控制估算基数（默认 4096，0=关闭），handler defer 释放，失败/中断/panic 均覆盖
- 调度：全部候选都在运行期冷却时，剩余冷却 ≤ OT_PROXY_COOLDOWN_GRACE 的最早恢复渠道
  兜底重试一次而不是 503（429 Retry-After 语义）；DB 层不可用的渠道不参与兜底
- 指标：openteam_proxy_requests_total / request_duration / channel_failures /
  usage_queue_depth / frozen_balance，/metrics 默认关闭（OT_METRICS_ENABLED 开启）

server/internal/pkg/metrics/metrics.go

@@ -0,0 +1,62 @@
+// Package metrics Prometheus 指标：进程内 registry + 文本端点（api 层按开关挂载 /metrics）。
+//
+// label 基数约束：channel/model 来自受管目录、status/kind 为有限枚举，可安全做 label；
+// 请求级随机值（trace id、用户 id、错误详情）禁止入 label，否则基数爆炸拖垮采集端。
+package metrics
+
+import (
+	"net/http"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/collectors"
+	"github.com/prometheus/client_golang/prometheus/promhttp"
+)
+
+var reg = prometheus.NewRegistry()
+
+var (
+	// RequestsTotal 代理请求终结计数。channel 为 "none" 表示未打到任何渠道
+	// （无候选/满载等），model 为请求的模型名。
+	RequestsTotal = prometheus.NewCounterVec(prometheus.CounterOpts{
+		Name: "openteam_proxy_requests_total",
+		Help: "Proxied requests terminated, by channel/model/status.",
+	}, []string{"channel", "model", "status"})
+
+	// RequestDuration 请求从进入 handler 到终结的耗时（按渠道）。
+	RequestDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
+		Name:    "openteam_proxy_request_duration_seconds",
+		Help:    "Proxy request duration, by channel.",
+		Buckets: []float64{.05, .1, .25, .5, 1, 2.5, 5, 10, 30, 60, 120, 300},
+	}, []string{"channel"})
+
+	// ChannelFailures 上游失败计数（按渠道与失败类型），与运行期冷却同源。
+	ChannelFailures = prometheus.NewCounterVec(prometheus.CounterOpts{
+		Name: "openteam_channel_failures_total",
+		Help: "Upstream failures, by channel/fail kind.",
+	}, []string{"channel", "kind"})
+
+	// RecorderQueueDepth 记账队列中待落库的记录数。持续增长说明 DB 写入跟不上。
+	RecorderQueueDepth = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "openteam_usage_queue_depth",
+		Help: "Usage records buffered, waiting to be flushed.",
+	})
+
+	// FrozenBalance 在途请求预扣（冻结）的金额合计（USD）。见 proxy 的 reserve/capture。
+	FrozenBalance = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "openteam_frozen_balance",
+		Help: "Balance reserved by in-flight requests (USD).",
+	})
+)
+
+func init() {
+	reg.MustRegister(
+		RequestsTotal, RequestDuration, ChannelFailures, RecorderQueueDepth, FrozenBalance,
+		collectors.NewGoCollector(),
+		collectors.NewProcessCollector(collectors.ProcessCollectorOpts{}),
+	)
+}
+
+// Handler /metrics 的 http.Handler（独立 registry，避免全局默认注册器的耦合）。
+func Handler() http.Handler {
+	return promhttp.HandlerFor(reg, promhttp.HandlerOpts{})
+}

server/internal/pkg/metrics/metrics_test.go

@@ -0,0 +1,58 @@
+package metrics
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+)
+
+// /metrics 文本端点：注入计数/观测后可被抓取，指标名与 label 齐全。
+func TestHandlerExposesInstruments(t *testing.T) {
+	RequestsTotal.WithLabelValues("c1", "m1", "success").Inc()
+	RequestsTotal.WithLabelValues("none", "m1", "error").Inc()
+	RequestDuration.WithLabelValues("c1").Observe(0.42)
+	ChannelFailures.WithLabelValues("c1", "rate_limited").Inc()
+	RecorderQueueDepth.Set(7)
+	FrozenBalance.Set(0.00246)
+
+	srv := httptest.NewServer(Handler())
+	defer srv.Close()
+	resp, err := http.Get(srv.URL)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		t.Fatalf("status = %d", resp.StatusCode)
+	}
+	buf := make([]byte, 1<<16)
+	n, _ := resp.Body.Read(buf)
+	body := string(buf[:n])
+	for _, want := range []string{
+		"openteam_proxy_requests_total{",
+		`channel="c1"`,
+		`status="success"`,
+		"openteam_proxy_request_duration_seconds",
+		"openteam_channel_failures_total",
+		"openteam_usage_queue_depth 7",
+		"openteam_frozen_balance 0.00246",
+		"go_goroutines", // Go 运行时指标一并暴露
+	} {
+		if !strings.Contains(body, want) {
+			t.Errorf("metrics 输出缺少 %q", want)
+		}
+	}
+	// 直方图观测应产生 bucket 样本
+	if !strings.Contains(body, `openteam_proxy_request_duration_seconds_bucket`) && n < len(buf) {
+		// 一次 Read 可能读不完，重读剩余部分再断言
+		rest := make([]byte, 1<<16)
+		n2, _ := resp.Body.Read(rest)
+		body += string(rest[:n2])
+		if !strings.Contains(body, "openteam_proxy_request_duration_seconds_bucket") {
+			t.Errorf("metrics 输出缺少直方图 bucket 样本")
+		}
+	}
+	_ = time.Second
+}