[ BOUNTY] Fix HighMemoryUsage self-dividing alert expression

[Xiaofei-Z]

Summary

Closes #1. Fixes the HighMemoryUsage Prometheus alert that divided process_resident_memory_bytes by itself (always evaluating to 1), making the alert unreliable.

Changes

• tools/monitoring_setup.py: replaced process_resident_memory_bytes / process_resident_memory_bytes > 0.9 with process_resident_memory_bytes / machine_memory_bytes > 0.9 — matches the existing instance:memory_usage:ratio recording rule already in the file
• Added validate_alert_expressions() regression guard that detects any self-dividing metric expression (metric / metric), runnable via the module's validation path

Testing

• python3 tools/monitoring_setup.py --validate → OK: no self-dividing alert expressions
• Reverse test: temporarily restoring the self-dividing expr → validator correctly flags it
• Confirmed machine_memory_bytes is valid PromQL and consistent with the recording rule at line 149
• python3 build.py run; diagnostic JSON attached (build-2b54872c.json)
• Existing dry-run output remains compatible

Checklist

• HighMemoryUsage no longer self-divides
• Replacement uses a meaningful denominator, valid PromQL
• Validation path catches self-dividing expressions
• Existing monitoring setup behavior preserved
• Diagnostic build log committed
• Changes scoped to monitoring setup path

[Xiaofei-Z]

Hi @weilixiong, submitted this for the bounty. Fixed the self-dividing expression (now uses machine_memory_bytes, consistent with the existing recording rule) and added a validator to catch this regression. Happy to address any feedback. Thanks!