feat: refine token usage display modes (#2329)

* feat: refine token usage display modes

* docs: clarify token usage accounting semantics

* fix: avoid duplicate subtask debug keys

* style: format token usage tests

* chore: address token attribution review feedback

* Update test_token_usage_middleware.py

* Update test_token_usage_middleware.py

* chore: simplify token attribution fallback

* fix token usage metadata follow-up handling

---------

Co-authored-by: Willem Jiang <willem.jiang@gmail.com>

frontend/src/core/settings/local.ts

@@ -1,9 +1,14 @@
+import type { TokenUsageInlineMode } from "../messages/usage-model";
 import type { AgentThreadContext } from "../threads";
 
 export const DEFAULT_LOCAL_SETTINGS: LocalSettings = {
   notification: {
     enabled: true,
   },
+  tokenUsage: {
+    headerTotal: true,
+    inlineMode: "per_turn",
+  },
   context: {
     model_name: undefined,
     mode: undefined,
@@ -22,6 +27,10 @@ export interface LocalSettings {
   notification: {
     enabled: boolean;
   };
+  tokenUsage: {
+    headerTotal: boolean;
+    inlineMode: TokenUsageInlineMode;
+  };
   context: Omit<
     AgentThreadContext,
     | "thread_id"
@@ -44,6 +53,10 @@ function mergeLocalSettings(settings?: Partial<LocalSettings>): LocalSettings {
       ...DEFAULT_LOCAL_SETTINGS.context,
       ...settings?.context,
     },
+    tokenUsage: {
+      ...DEFAULT_LOCAL_SETTINGS.tokenUsage,
+      ...settings?.tokenUsage,
+    },
     notification: {
       ...DEFAULT_LOCAL_SETTINGS.notification,
       ...settings?.notification,