feat(memtrack): record per-benchmark pid in execution timestamps

[GuillaumeLagrange]

Memory mode tracks the whole runner process tree, so between Start/StopProfiler the harness parent and sibling worker processes emit allocation events into every measured region. The CurrentBenchmark FIFO command already carries the pid that ran each benchmark; capture it alongside the uri in ExecutionTimestamps.bench_pid_by_ts (serde-default for artifacts written before this field) so the parser can scope events to the benchmarking process.

[greptile-apps]

Greptile Summary

This PR records benchmark PIDs alongside execution timestamps and adds fork events for memtrack process-tree filtering. The main changes are:

• A new bench_pid_by_ts field on execution timestamp artifacts.
• FIFO handling that stores each CurrentBenchmark PID with its timestamp.
• A new memtrack Fork event type in the eBPF event schema.
• Rust parsing and serialization support for fork events.

Confidence Score: 4/5

This is close, but the fork-event gap should be fixed before merging.

• Fork edges can be omitted while PID tracking still records the child.
• The new descendant filtering can then drop allocations from a tracked child process.
• The schema and parser changes otherwise look consistent in the changed code.

crates/memtrack/src/ebpf/c/memtrack.bpf.c

Important Files Changed

Filename	Overview
crates/memtrack/src/ebpf/c/memtrack.bpf.c	Adds fork-event emission, but the profiling gate can omit edges needed for descendant filtering.
crates/memtrack/src/ebpf/c/event.h	Adds the fork event type and parent-PID payload.
crates/memtrack/src/ebpf/events.rs	Parses fork events into the shared memtrack event model.
crates/runner-shared/src/artifacts/execution_timestamps.rs	Adds benchmark PID timestamps with defaulting for older artifacts.
crates/runner-shared/src/artifacts/memtrack/mod.rs	Adds the shared Fork event variant and serialization coverage.
src/executor/shared/fifo.rs	Stores the PID from CurrentBenchmark beside the benchmark URI timestamp.

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
crates/memtrack/src/ebpf/c/memtrack.bpf.c:121
**Fork Edges Can Disappear**

This gate can drop the process-tree edge that the new PID scoping depends on. The handler still adds the child to `tracked_pids` and `pids_ppid` before this branch, so a child forked while profiling is disabled can later emit allocation events after profiling starts, but the artifact never contains the matching `Fork` event. When the parser scopes events to the benchmark PID and descendants, that child has no recorded parent edge and its allocations can be treated as unrelated and dropped.

_{Reviews (5): Last reviewed commit: "feat(memtrack): record per-benchmark pid..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 17 untouched benchmarks

---

_{Comparing cod-3010-improve-node-memory-benchmarks-stability (663ead6) with cod-1801-only-attach-to-used-libraries-in-memtrack (4d3fbb0)}

[greptile-apps]

Fork Edges Can Disappear

This gate can drop the process-tree edge that the new PID scoping depends on. The handler still adds the child to tracked_pids and pids_ppid before this branch, so a child forked while profiling is disabled can later emit allocation events after profiling starts, but the artifact never contains the matching Fork event. When the parser scopes events to the benchmark PID and descendants, that child has no recorded parent edge and its allocations can be treated as unrelated and dropped.

Prompt To Fix With AI

This is a comment left during a code review.
Path: crates/memtrack/src/ebpf/c/memtrack.bpf.c
Line: 121

Comment:
**Fork Edges Can Disappear**

This gate can drop the process-tree edge that the new PID scoping depends on. The handler still adds the child to `tracked_pids` and `pids_ppid` before this branch, so a child forked while profiling is disabled can later emit allocation events after profiling starts, but the artifact never contains the matching `Fork` event. When the parser scopes events to the benchmark PID and descendants, that child has no recorded parent edge and its allocations can be treated as unrelated and dropped.

How can I resolve this? If you propose a fix, please make it concise.

[not-matthias]

LGTM overall

[not-matthias]

I dont think we should do this type of filtering here, because it's still useful to be able to track the full execution tree. We have the timestamp so we can do the processing in the backend

[not-matthias]

e.g. for RSS tracking I also needed to track fork/exec/exit even when memtrack isnt enabled: 8ef53f4 (#453)