[CI] Use L2 ccache reduce build time in all workflows

[SigureMo]

PR Category

Execute Infrastructure

PR Types

Devs

Description

基于 #77167 的思路尝试优化全流水线构建架构，使用机内 L1 本地 cache + 机间 L2 CFS cache 来降低构建时间，每台机器约占用 50G 存储即可（跨流水线共用，只需要 50G）

原有架构：

新架构：

coverage 首次编译 5h30m：

Cacheable calls:   4750 / 4750 (100.0%)
  Hits:               0 / 4750 ( 0.00%)
    Direct:           0
    Preprocessed:     0
  Misses:          4750 / 4750 (100.0%)
Local storage:
  Cache size (GB):  7.6 / 50.0 (15.26%)
  Hits:               0 / 4750 ( 0.00%)
  Misses:          4750 / 4750 (100.0%)
Remote storage:
  Hits:               0 / 4750 ( 0.00%)
  Misses:          4750 / 4750 (100.0%)

coverage 重复编译 38min，低于 L1 CFS 的 55min：

Cacheable calls:   4750 / 4750 (100.0%)
  Hits:            4744 / 4750 (99.87%)
    Direct:        4744 / 4744 (100.0%)
    Preprocessed:     0 / 4744 ( 0.00%)
  Misses:             6 / 4750 ( 0.13%)
Local storage:
  Cache size (GB):  7.7 / 50.0 (15.39%)
  Hits:            4744 / 4750 (99.87%)
  Misses:             6 / 4750 ( 0.13%)
Remote storage:
  Hits:               0 /    6 ( 0.00%)
  Misses:             6 /    6 (100.0%)

TODOs

• 添加流水线定期清理 L2 ccache

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[Copilot]

Pull request overview

This PR optimizes the CI build pipeline by implementing a two-tier ccache architecture: L1 local cache on individual machines (50GB) and L2 shared cache on CFS (Common File System) for cross-machine sharing.

Key changes:

• Migrated from single-tier ccache (200GB) to L1 (50GB local) + L2 (shared CFS) architecture
• Updated runner group from GZ_BD-CPU to coverage-build-l2-debug for dedicated L2-optimized infrastructure
• Modified Docker volume mounts to support the new cache directory structure

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.