docs: link PR#3006 (microcompaction) to P0 core#1 (multi-layer compression)

PR#3006 implements L2 idle microcompaction, which is one layer of the
5-layer compression system described in core#1. Previously only tracked
under P2 perf#13 (cache_edits), now also linked to the P0 parent item.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: wenshao

docs/comparison/qwen-code-improvement-report.md

@@ -40,7 +40,7 @@
 | 优先级 | 改进点 | Qwen Code 现状 | 难度 | 进展 |
 |:------:|--------|----------------|:----:|------|
 | **P0** | [[Mid-Turn Queue Drain](./command-queue-orchestration-deep-dive.md)](./input-queue-deep-dive.md) — Agent 执行中途注入用户输入，无需等整轮结束 [↓](./qwen-code-improvement-report-p0-p1-core.md#item-6) | 推理循环内无队列检查 | 中 | [PR#2854](https://github.com/QwenLM/qwen-code/pull/2854) ✓ |
-| **P0** | [多层上下文压缩](./context-compression-deep-dive.md) — 自动裁剪旧工具结果 + 摘要，用户无需手动 /compress [↓](./qwen-code-improvement-report-p0-p1-core.md#item-1) | 仅单一 70% 手动压缩 | 中 | — |
+| **P0** | [多层上下文压缩](./context-compression-deep-dive.md) — 自动裁剪旧工具结果 + 摘要，用户无需手动 /compress [↓](./qwen-code-improvement-report-p0-p1-core.md#item-1) | 仅单一 70% 手动压缩 | 中 | [PR#3006](https://github.com/QwenLM/qwen-code/pull/3006)（L2 microcompaction） |
 | **P0** | [Fork Subagent](./fork-subagent-deep-dive.md) — Subagent 继承完整对话上下文，共享 prompt cache 省 80%+ 费用 [↓](./qwen-code-improvement-report-p0-p1-core.md#item-2) | Subagent 必须从零开始 | 中 | [PR#2936](https://github.com/QwenLM/qwen-code/pull/2936) |
 | **P0** | [会话崩溃恢复与中断检测](./crash-recovery-deep-dive.md) — 3 种中断状态检测 + 合成续行 + 全量恢复 [↓](./qwen-code-improvement-report-p0-p1-engine.md#item-7) | 无崩溃恢复 | 大 | — |
 | **P1** | [Speculation](../tools/claude-code/10-prompt-suggestions.md) — 预测用户下一步并提前执行，Tab 接受零延迟 [↓](./qwen-code-improvement-report-p0-p1-core.md#item-3) | 已实现但默认关闭 | 小 | [PR#2525](https://github.com/QwenLM/qwen-code/pull/2525) ✓ |