Fix Claude agent interrupt race and refactor mode color tiers

- Fix race in agent-manager where pendingRun wasn't fully cleaned up
  before the next streamAgent call, causing "already has an active run"
- Fix Claude agent query restart: null out query/input before awaiting
  old iterator return so the old pump skips failActiveTurns
- Reset pendingInterruptAbort on new foreground turn
- Add system error assertion to send-during-tool-call e2e test
- Rename mode color tiers: drop "default", rename "readonly" to
  "planning", update color assignments across providers and UI

packages/server/src/server/agent/agent-manager.ts

@@ -1383,6 +1383,13 @@ export class AgentManager {
           timeout,
         ]);
       }
+      // The waiter settling wakes up the streamForwarder generator, but its
+      // finally block (which deletes the pendingForegroundRun) runs asynchronously.
+      // Wait for the pending run to be fully cleaned up so the next streamAgent
+      // call doesn't see a stale entry and reject with "already has an active run".
+      if (pendingRun && !pendingRun.settled) {
+        await Promise.race([pendingRun.settledPromise, timeout]);
+      }
     } else if (pendingRun) {
       const timeout = new Promise<void>((resolve) => setTimeout(resolve, 2000));
       await Promise.race([pendingRun.settledPromise, timeout]);