fix(coding-agent): respected google retry hints

Parsed provider retry hint text before falling back to quota cooldowns so Google per-minute token limits use the server-provided delay.

Fixes #1253

packages/coding-agent/src/session/agent-session.ts

@@ -76,6 +76,7 @@ import {
 } from "@oh-my-pi/pi-ai";
 import { MacOSPowerAssertion } from "@oh-my-pi/pi-natives";
 import {
+	extractRetryHint,
 	getAgentDbPath,
 	isEnoent,
 	isUnexpectedSocketCloseMessage,
@@ -6949,6 +6950,11 @@ export class AgentSession {
 			}
 		}
 
+		const retryHintMs = extractRetryHint(undefined, errorMessage);
+		if (retryHintMs !== undefined) {
+			return retryHintMs;
+		}
+
 		const resetMsMatch = /x-ratelimit-reset-ms\s*[:=]\s*(\d+)/i.exec(errorMessage);
 		if (resetMsMatch) {
 			const resetMs = Number(resetMsMatch[1]);