Fix HighMemoryUsage alert ratio

[mirzakah]

Summary

Refs #1. Fixes the recommended HighMemoryUsage Prometheus alert so it no longer divides process_resident_memory_bytes by itself. The rule now compares process resident memory to machine_memory_bytes using on(instance) vector matching while keeping the existing threshold and alert metadata.

Changes

• Replaced the self-dividing HighMemoryUsage expression with a real memory usage ratio.
• Added a stdlib smoke test that can run directly with Python or under pytest and catches this regression.
• Included diagnostic/build-8c3fd31c.json from the required python3 build.py run.

Testing

• python3 tools/test_monitoring_setup_memory_alert.py passed.
• /Users/mirzakahrovic/Documents/codex-bounty-grinder/.venv/bin/python -m pytest tools/test_monitoring_setup_memory_alert.py -q passed.
• python3 -m py_compile tools/monitoring_setup.py tools/test_monitoring_setup_memory_alert.py passed.
• git diff --check HEAD~1..HEAD passed.
• python3 build.py ran and generated diagnostic/build-8c3fd31c.json; 3/10 modules passed (frontend, market, v2-market-stream). The remaining failures are local toolchain/environment failures (cargo, cmake, Java runtime, luac, ghc) plus the existing macOS frailbox MAP_HUGETLB compile issue.
• .logd creation failed locally with encryptly error b400a134ad13c8d7c8e3; the diagnostic JSON records this failure.

Checklist

• Relevant Python tool smoke test passes locally
• Tests pass locally
• Diagnostic build log is committed in this PR (.logd creation failed locally; diagnostic JSON is committed)
• Documentation has been updated, if applicable
• Configuration or schema changes are documented, if applicable
• No generated build artifacts are committed, except the diagnostic build metadata
• Changes are scoped to the PR purpose and avoid unrelated cleanup
• Security, privacy, and error-handling implications have been considered

---

• I would like to request that my diagnostic build log is removed before merging