fix: 统一跨 Peer 时延指标的标签与秒分桶

Author: xiaoland

docs/30-unit-tdd/client-runtime-and-delegation.md

@@ -151,7 +151,9 @@ exporter 为 10 秒，经 relay 为 35 秒，以覆盖服务端最多 32 秒的
 `JobManager.create` 在同一个 PostgREST INSERT 中写入 SDK 注入的提交 carrier。超出
 512 UTF-8 bytes 的可选 tracestate 会被省略并计数。执行成功 claim 后创建独立 trace，
 通过 SDK Span Link 关联提交；claim、取消和 close 不重写提交列。Job ID 是诊断关联字段，
-不是 trace ID。`job.submitted`、`job.started`、`job.closed` 只从明确的业务边界发出结构化
+不是 trace ID。时延指标 `inkcre.operation.duration` 使用与 Core 一致的 `operation` /
+`outcome` 标签和 5ms 至 300s 的显式秒分桶，便于跨 Peer 聚合；日志保留 `inkcre.*` 属性。
+`job.submitted`、`job.started`、`job.closed` 只从明确的业务边界发出结构化
 事件；既有应用日志、异常 message、参数和内容不会桥接到新增 OTLP。
 
 浏览器原生 `await` 不能依赖 StackContextManager 或 ZoneContextManager 自动保留当前

packages/core/src/obsrv/telemetry.ts

@@ -177,11 +177,23 @@ export async function configureTelemetry(
     logger = provider.getLogger('inkcre.browser')
   }
   if (endpoints.metrics) {
-    const { MeterProvider, PeriodicExportingMetricReader } =
+    const { MeterProvider, PeriodicExportingMetricReader, AggregationType } =
       await import('@opentelemetry/sdk-metrics')
     const { OTLPMetricExporter } = await import('@opentelemetry/exporter-metrics-otlp-proto')
     const provider = new MeterProvider({
       resource,
+      // Match the Core histogram so cross-Peer aggregation keeps seconds and buckets aligned.
+      views: [
+        {
+          instrumentName: 'inkcre.operation.duration',
+          aggregation: {
+            type: AggregationType.EXPLICIT_BUCKET_HISTOGRAM,
+            options: {
+              boundaries: [0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1, 2.5, 5, 10, 30, 60, 300],
+            },
+          },
+        },
+      ],
       readers: [
         new PeriodicExportingMetricReader({
           exporter: new OTLPMetricExporter({
@@ -255,7 +267,10 @@ export async function observeOperation<T>(
   } finally {
     span.end()
     const attributes = { 'inkcre.operation': name, 'inkcre.outcome': result.outcome }
-    operationDuration?.record((performance.now() - started) / 1000, attributes)
+    operationDuration?.record((performance.now() - started) / 1000, {
+      operation: name,
+      outcome: result.outcome,
+    })
     operationLogger?.emit({
       context: active,
       eventName: 'inkcre.operation.completed',