docs: 合并性能分析报告并优化内存管理

将 performance-reporter.md 合入 memory-peak-analysis.md，统一分析文档。
代码优化包括：compact 峰值释放、GC 阈值触发、虚拟滚动参数调优、
HybridTransport 队列缩减、无界缓存加 LRU 淘汰、taskSummary 避免数组拷贝。

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

src/cli/print.ts

@@ -551,9 +551,19 @@ export async function runHeadless(
     proactiveModule.activateProactive('command')
   }
 
-  // Periodically force a full GC to keep memory usage in check
+  // Periodically run GC to keep memory usage in check.
+  // Uses a memory threshold to trigger a forced (major) GC when RSS grows
+  // beyond 350MB — the incremental GC may not reclaim enough during peaks
+  // (compact, long sessions with many mounted DOM nodes).
   if (typeof Bun !== 'undefined') {
-    const gcTimer = setInterval(Bun.gc, 1000)
+    const gcTimer = setInterval(() => {
+      const rss = process.memoryUsage.rss()
+      if (rss > 350 * 1024 * 1024) {
+        Bun.gc(true)
+      } else {
+        Bun.gc(false)
+      }
+    }, 1000)
     gcTimer.unref()
   }
 

src/cli/transports/HybridTransport.ts

@@ -83,7 +83,7 @@ export class HybridTransport extends WebSocketTransport {
       // SerialBatchEventUploader backpressure check). So set it high enough
       // to be a memory bound only. Wire real backpressure in a follow-up
       // once callers await.
-      maxQueueSize: 100_000,
+      maxQueueSize: 10_000,
       baseDelayMs: 500,
       maxDelayMs: 8000,
       jitterMs: 1000,