fix(inference): preserve legacy and localized tooltips

Keep an explicit legacy offload fallback when detailed runtime metadata is unavailable, and rebuild unofficial overlay layers when the locale changes.

中文：保留旧版及本地化提示卡。在缺少详细运行时元数据时显示明确的旧版卸载状态，并在语言切换后重新构建非正式运行覆盖层。

Author: cquil11

packages/app/src/components/inference/ui/ScatterGraph.tsx

@@ -2283,6 +2283,7 @@ const ScatterGraph = React.memo(
       yLabel,
       selectedYAxisMetric,
       chartDefinition,
+      locale,
     ]);
 
     // Layers handle for the decoration effect — lets it re-run individual

packages/app/src/components/inference/utils/tooltip-utils.test.ts

@@ -262,6 +262,21 @@ describe('generateTooltipContent', () => {
     expect(html).not.toContain('Offload Mode');
   });
 
+  it('keeps a clearly marked binary fallback for legacy agentic rows', () => {
+    const enabled = generateTooltipContent(
+      tooltipConfig({ data: pt({ offload_mode: 'on', kv_offloading: undefined }) }),
+    );
+    const disabledZh = generateTooltipContent(
+      tooltipConfig({
+        locale: 'zh',
+        data: pt({ offload_mode: 'off', kv_offloading: undefined }),
+      }),
+    );
+
+    expect(enabled).toContain('<strong>Offload Type:</strong> Enabled (legacy data)');
+    expect(disabledZh).toContain('<strong>卸载类型:</strong> 已禁用（旧版数据）');
+  });
+
   it('shows multinode KV transfer and cache-hit metadata for fixed-sequence points', () => {
     const html = generateTooltipContent(
       tooltipConfig({

packages/app/src/components/inference/utils/tooltipUtils.ts

@@ -103,6 +103,8 @@ const CACHE_STRINGS = {
     cpuHitRate: 'CPU Cache Hit Rate',
     theoreticalHitRate: 'Theoretical Cache Hit Rate',
     none: 'None',
+    legacyEnabled: 'Enabled (legacy data)',
+    legacyDisabled: 'Disabled (legacy data)',
   },
   zh: {
     offloadType: '卸载类型',
@@ -113,6 +115,8 @@ const CACHE_STRINGS = {
     cpuHitRate: 'CPU Cache 命中率',
     theoreticalHitRate: '理论 Cache 命中率',
     none: '无',
+    legacyEnabled: '已启用（旧版数据）',
+    legacyDisabled: '已禁用（旧版数据）',
   },
 } as const;
 
@@ -149,6 +153,9 @@ const generateCacheMetadataHTML = (d: InferenceData, locale: Locale): string =>
   const parts: string[] = [];
   if (d.kv_offloading) {
     parts.push(tooltipLine(t.offloadType, offloadTypeLabel(d.kv_offloading, locale)));
+  } else if (d.offload_mode) {
+    const enabled = d.offload_mode.toLowerCase() === 'on';
+    parts.push(tooltipLine(t.offloadType, enabled ? t.legacyEnabled : t.legacyDisabled));
   }
   if (d.kv_offload_backend) {
     const backend = cacheImplementationLabel(d.kv_offload_backend);