test(runtime): add scheduler service latency benchmark

[Coldwings]

Description

Adds a focused mixed-workload benchmark for scheduler service latency. It measures how long a single worker takes to service an already-armed I/O completion and a remote MPSC-inbox submission while a yield loop keeps the local runnable deque non-empty.

The benchmark reports p50, p99, and maximum latency, records the selected I/O backend, refuses to mix samples from different backends, and bounds every setup/completion/shutdown phase. Its --smoke mode is added to Release CI only as build-and-termination coverage; noisy shared-runner timing values are deliberately not pass/fail thresholds.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Performance improvement (optimization that improves speed/memory usage)
• Documentation (changes to documentation, comments, or examples)
• Refactoring (code changes that neither fix bugs nor add features)
• Tests (adding or modifying tests)
• Build/CI (changes to build system, CI configuration, or dependencies)

Related Issues

Closes #1015

Changes Made

Core Changes

• Added scheduler_service_benchmark with configurable trial count and backlog duration.
• Added a short CI smoke invocation without timing thresholds.
• Documented the benchmark, measurement semantics, and interpretation guidance in README, wiki, and changelog.

API Changes

None.

Testing

Unit Tests

• Added benchmark termination coverage in CI
• Existing scheduler tests pass locally
• Existing I/O tests pass locally

Sanitizer Testing

• Instrumented benchmark smoke passed with ASAN and leak detection
• Instrumented benchmark smoke passed with TSAN
• No sanitizer reports

Test Results

Release build with developer warnings and -Werror:
  scheduler_service_benchmark built successfully

scheduler_service_benchmark --smoke:
  3/3 trials completed

scheduler_service_benchmark (40 trials, 20 ms backlog, io_uring):
  I/O p50=20.08 ms, p99=20.17 ms
  remote inbox p50=20.08 ms, p99=20.16 ms

elio_tests "[scheduler]":
  408 assertions in 73 test cases passed

elio_tests "[io]":
  699 assertions in 75 test cases passed

ASAN smoke: passed with leak detection
TSAN smoke: passed with no race report

GCC emits its existing atomic_thread_fence is not supported with -fsanitize=thread warnings from runtime headers, so the standalone TSAN benchmark build disabled warnings-as-errors while retaining TSAN instrumentation and halt-on-error behavior.

Checklist

Code Quality

• Code follows the project style
• No debug code, TODOs, or commented-out code
• Release build produces no new warnings under -Werror

Documentation

• Updated README, wiki, and changelog
• Documented benchmark options and result interpretation

Testing

• Added coverage that proves the benchmark builds and terminates
• Relevant existing tests pass locally
• Tested with ASAN and TSAN

Compatibility

• Backward compatible; no public API change
• Linux-only eventfd dependency matches Elio's supported platform

Performance

• Measurement does not impose unstable timing thresholds in CI
• Baseline captures the intended current scheduler service delay

Reviewer Guidance

Please focus on:

• whether the armed-I/O and remote-inbox setup isolates scheduler service delay correctly;
• whether release/acquire publication and failure cleanup keep all task state alive until shutdown;
• whether the smoke mode is appropriately bounded and non-flaky across io_uring and epoll fallback environments.

[Copilot]

Pull request overview

Adds a focused scheduler service-latency benchmark for armed I/O and remote inbox work under sustained runnable load.

Changes:

• Adds configurable trials, smoke mode, percentile reporting, and bounded cleanup.
• Integrates a Release CI smoke run.
• Documents usage and measurement interpretation.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
examples/scheduler_service_benchmark.cpp	Implements the benchmark.
examples/CMakeLists.txt	Adds the benchmark target.
.github/workflows/ci.yml	Runs smoke coverage in Release CI.
README.md	Lists benchmark build and run commands.
wiki/Performance-Tuning.md	Documents interpretation guidance.
CHANGELOG.md	Records the new benchmark.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.