✨ feat: Add model performance metrics to dashboard

Add a shared `performance-metrics` feature module for perf metric APIs, DTOs, and formatting, then surface global 24h model performance on the dashboard with cards and a top-model table.

Reuse the shared metrics module from pricing model details, remove duplicated perf API/formatting code from pricing, and add localized labels for the new dashboard performance UI.

web/default/src/features/performance-metrics/api.ts

@@ -0,0 +1,24 @@
+import { api } from '@/lib/api'
+import type { PerformanceMetricsData, PerfSummaryAllData } from './types'
+
+export async function getPerfMetricsSummary(
+  hours = 24
+): Promise<PerfSummaryAllData> {
+  const res = await api.get<PerfSummaryAllData>('/api/perf-metrics/summary', {
+    params: { hours },
+  })
+  return res.data
+}
+
+export async function getPerfMetrics(
+  modelName: string,
+  hours = 24
+): Promise<PerformanceMetricsData> {
+  const res = await api.get<PerformanceMetricsData>('/api/perf-metrics', {
+    params: {
+      model: modelName,
+      hours,
+    },
+  })
+  return res.data
+}

web/default/src/features/performance-metrics/lib/format.ts

@@ -0,0 +1,16 @@
+export function formatThroughput(tps: number): string {
+  if (tps <= 0) return '—'
+  if (tps >= 1_000) return `${(tps / 1_000).toFixed(1)}K t/s`
+  return `${tps.toFixed(tps < 10 ? 2 : 1)} t/s`
+}
+
+export function formatLatency(ms: number): string {
+  if (!Number.isFinite(ms) || ms <= 0) return '—'
+  if (ms >= 1_000) return `${(ms / 1_000).toFixed(2)}s`
+  return `${Math.round(ms)}ms`
+}
+
+export function formatUptimePct(pct: number): string {
+  if (!Number.isFinite(pct)) return '—'
+  return `${pct.toFixed(2)}%`
+}

web/default/src/features/performance-metrics/types.ts

@@ -0,0 +1,42 @@
+export type PerformanceSeriesPoint = {
+  ts: number
+  avg_ttft_ms: number
+  avg_latency_ms: number
+  success_rate: number
+  avg_tps: number
+}
+
+export type PerformanceGroup = {
+  group: string
+  avg_ttft_ms: number
+  avg_latency_ms: number
+  success_rate: number
+  avg_tps: number
+  series: PerformanceSeriesPoint[]
+}
+
+export type PerformanceMetricsData = {
+  success: boolean
+  message?: string
+  data: {
+    model_name: string
+    series_schema?: string
+    groups: PerformanceGroup[]
+  }
+}
+
+export type PerfModelSummary = {
+  model_name: string
+  avg_latency_ms: number
+  success_rate: number
+  avg_tps: number
+  request_count: number
+}
+
+export type PerfSummaryAllData = {
+  success: boolean
+  message?: string
+  data: {
+    models: PerfModelSummary[]
+  }
+}