AGC Rust 分片改为失败用例有界复核

- run-rust-shell-test-shards.mjs 在分片失败后解析失败用例名，仅对这组用例在新进程内复核一次
- 复核通过按 flaky 明确记录并通过，复核仍失败则保留失败并标注 re-run 结果，避免洗掉真实回归
- 解析兼容 runner 输出的 [rust-shards] 前缀与 panic 头格式，已用一次性探针 crate 端到端验证重跑路径

apps/ai-game-creator-shell/scripts/run-rust-shell-test-shards.mjs

@@ -371,6 +371,26 @@ async function runWithConcurrency(shards, runner) {
   return results;
 }
 
+function extractFailingTestNames(result) {
+  const names = new Set();
+  for (const line of result.failures) {
+    const normalized = line
+      .trim()
+      .replace(/^\[rust-shards\]\s*/, '')
+      .replace(/^----\s*/, '')
+      .replace(/\s*stdout\s*----$/, '')
+      .replace(/\s*\(\d+\)\s*$/, '')
+      .trim();
+    const match = normalized.match(
+      /^(process_session::tests::[A-Za-z0-9_:]+|tests::[A-Za-z0-9_:]+)$/,
+    );
+    if (match) {
+      names.add(match[1]);
+    }
+  }
+  return [...names];
+}
+
 async function main() {
   const executable = await resolveTestExecutable();
   const testNames = await listTestNames(executable);
@@ -410,15 +430,55 @@ async function main() {
 
   const results = await runWithConcurrency(
     selectedShards,
-    ({ index, shardTestNames }) =>
-      runShard(executable, index, shards.length, shardTestNames),
+    async ({ index, shardTestNames }) => {
+      const result = await runShard(
+        executable,
+        index,
+        shards.length,
+        shardTestNames,
+      );
+      if (result.ok) {
+        return result;
+      }
+      // 片内串行的时序型用例在高负载 CI 上会偶发假红。只对失败用例做一次
+      // 有界复核：复核通过按 flaky 记录，复核失败才判红，避免把真实回归洗掉。
+      const failingTestNames = extractFailingTestNames(result).filter((name) =>
+        shardTestNames.includes(name),
+      );
+      if (failingTestNames.length === 0) {
+        return result;
+      }
+      const retry = await runShard(
+        executable,
+        index,
+        shards.length,
+        failingTestNames,
+      );
+      if (!retry.ok) {
+        return {
+          ...result,
+          failures: [
+            `re-run of ${failingTestNames.length} failing test(s) also failed`,
+            ...retry.failures,
+          ],
+        };
+      }
+      return {
+        ...result,
+        ok: true,
+        retriedTestNames: failingTestNames,
+      };
+    },
   );
 
   let failed = false;
   for (const result of results) {
     if (result.ok) {
+      const retrySuffix = result.retriedTestNames
+        ? ` (flaky: re-ran ${result.retriedTestNames.length} failing test(s) and passed: ${result.retriedTestNames.join(', ')})`
+        : '';
       console.log(
-        `[rust-shards] ${result.label} ok: ${result.testCount} test(s) in ${formatDuration(result.durationMs)}`,
+        `[rust-shards] ${result.label} ok: ${result.testCount} test(s) in ${formatDuration(result.durationMs)}${retrySuffix}`,
       );
       continue;
     }