fix(linux): bound the foreground process-tree scan

[caner-akca]

Summary

• Fix the headless server's read-syscall rate (and CPU) growing without bound over uptime on Linux at a constant pane count: foreground-job detection walked each pane shell's entire descendant process tree and read /proc/<pid>/stat per node, so accumulated child/zombie processes made every probe read more over time.
• Bound the primary foreground process-tree traversal (process_tree_pids in src/platform/linux.rs) with a 512-node scan limit, truncating the walk while retaining the enqueued shell/foreground-group roots, so the detected agent and normal foreground groups are unchanged and only pathologically large trees are capped. Mirrors the existing CHILD_GROUPS_SCAN_LIMIT on the fallback path.

Validation

• live: stock 0.8.2 at /home/exedev/herdr-fix-3621/stock/target/release/herdr (sha256:b52c743d51046f9173761213b4f05063d7f5c1334753c7e89dfd6e8a71e7452e) showed the server's /proc/<pid>/io read rate scale with foreground-tree size (18,975/s at 700 descendants → 54,124/s at 2,100, ×2.85); frozen driver sha256:7439146b9ba31d48b921a1e4c6bfdb84c5b7cdab548d69680553f3ef88f85c59 with HERDR_BIN=/home/exedev/herdr-fix-3621/patched/target/release/herdr (sha256:6f498481b8578918b30346fca1dfc9079aefa082b9b37152f0f4504558a57613) did not (14,519/s → 13,925/s, ×0.95, bounded)
• regression coverage: foreground_tree_traversal_is_bounded_by_the_scan_limit failed without the fix and passes at the committed HEAD
• focused: cargo nextest run -E 'test(foreground_tree_traversal_is_bounded_by_the_scan_limit)' passed at the committed HEAD
• just check passed at the committed HEAD

Notes

Reproduced and validated on a Linux x86_64 host (kernel 6.12, 2 cores). The 100%-/proc/<pid>/stat read attribution and tree-size scaling were confirmed with strace; the accelerated driver grows two pane shells' descendant trees (both above the 512-node bound) and compares the server read rate, standing in for the reporter's real agents accumulating descendants over ~20h.

refs #3621

[coderabbitai]

Important

Review skipped

Auto reviews are limited based on label configuration.

🏷️ Required labels (at least one) (1)

• ai-review

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Team

Run ID: c8bffb29-52a5-4f04-9287-1402339de6fe

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}