fix(delegation): replace message-count-stability polling with native finish-based completion detection

Sync task completion was fragile — detecting premature stability during
brief idle periods between tool calls. Now mirrors opencode's native
SessionPrompt.loop() logic: checks assistant finish reason is terminal
(not tool-calls/unknown) and assistant.id > user.id.

Also switches sync prompt sender from blocking HTTP (promptSync) to
async fire-and-forget (promptAsync) to avoid JSON parse errors in ACP.

src/tools/delegate-task/sync-prompt-sender.ts

@@ -1,6 +1,6 @@
 import type { DelegateTaskArgs, OpencodeClient } from "./types"
 import { isPlanFamily } from "./constants"
-import { promptSyncWithModelSuggestionRetry } from "../../shared/model-suggestion-retry"
+import { promptWithModelSuggestionRetry } from "../../shared/model-suggestion-retry"
 import { formatDetailedError } from "./error-formatting"
 
 export async function sendSyncPrompt(
@@ -17,7 +17,7 @@ export async function sendSyncPrompt(
 ): Promise<string | null> {
   try {
     const allowTask = isPlanFamily(input.agentToUse)
-    await promptSyncWithModelSuggestionRetry(client, {
+    await promptWithModelSuggestionRetry(client, {
       path: { id: input.sessionID },
       body: {
         agent: input.agentToUse,