fix: apply cpu affinity to existing agent threads

[kylewanginchina]

This PR is for:

• Agent

Fixes CPU affinity updates not being applied to already running agent threads, and replaces manual procfs thread scanning with the procfs library

Steps to reproduce the bug

• Start a managed deepflow-agent and let it create its normal runtime worker threads.
• Update global.tunning.cpu_affinity from deepflow-server for the agent group after those threads are already running.
• Observe that the new CPU affinity setting is expected to apply to all existing agent threads, not only to threads created later.
• In the original implementation, the rebinding logic depended on manual reads from /proc/self/task and /proc/self/task/<tid>/comm.
• This made the thread scanning code harder to maintain and duplicated functionality already provided by the procfs crate.

Changes to fix the bug

• Apply CPU affinity updates to already existing agent threads when global.tunning.cpu_affinity changes, instead of limiting the effect to threads
  created after startup.
• Keep scanning multiple times so newly discovered threads created during the rebinding window can also be covered.
• Continue skipping self-managed kick-kern.* eBPF threads during CPU affinity rebinding.
• Replace manual /proc/self/task directory traversal with procfs::process::Process::myself()?.tasks()?.
• Replace direct reads of /proc/self/task/<tid>/comm with task.status()?.name.
• Add a small procfs::ProcError to io::Error conversion helper so the existing retry and error-handling flow remains unchanged.
• Preserve the original tolerance for disappearing threads by continuing on NotFound errors during task enumeration and thread name resolution.

Affected branches

• main

Checklist

• Added unit test to verify the fix.
• Verified eBPF program runs successfully on linux 4.14.x.
• Verified eBPF program runs successfully on linux 4.19.x.
• Verified eBPF program runs successfully on linux 5.2.x.
• Verified eBPF program runs successfully on linux 5.4.x.

[kylewanginchina]

针对这个工单需求

[kylewanginchina]

测试结果证明

• 本地单元测试

  • 命令：

    docker exec deepflow-ext-ide /bin/bash -c "cd /work/deepflow/agent && cargo test --lib config::handler::tests -- --test-threads=1"

  • 结果：6 passed; 0 failed

  • 关键输出：

    running 6 tests
    test config::handler::tests::test_set_cpu_affinity_updates_existing_worker_threads ... ok
    test result: ok. 6 passed; 0 failed

• 格式检查

  • 命令：

    docker exec deepflow-ext-ide /bin/bash -c "cd /work/deepflow/agent && cargo fmt --check"

  • 结果：退出码 0

• 10.50.120.200 managed 模式实机验证

  • 我在 10.50.120.201 上临时给目标 agent group profiling 下发了：

    global:
    tunning:
    cpu_affinity:
    - 0

  • 然后在 10.50.120.200 用 managed 模式启动：

    /home/deepflow-agent -f /home/deepflow-agent.yaml

  • 日志证据：

    • 10.50.120.200:/var/log/deepflow-agent/deepflow-agent.log:4367

    • 关键行：

      Update global.tunning.cpu_affinity from [] to [0].

  • 线程亲和性实测输出里，受控模式进程 PID=3997103 的多个工作线程都已收敛到 CPU 0

    • 示例：

      3997103: pid 3997103's current affinity list: 0
      3997148: pid 3997148's current affinity list: 0
      3997150: pid 3997150's current affinity list: 0
      3997152: pid 3997152's current affinity list: 0
      3997157: pid 3997157's current affinity list: 0

    • 覆盖到的线程包含 uniform-sender、stats-collector、main-loop、tokio-runtime-w、guard 等，不是只有主线程

[kylewanginchina]

测试结果：