feat: Resource-aware job scheduling

[nigel-dev]

Source

Audit report — Section 9: Phase 3 Roadmap

Description

MC currently enforces maxParallel as a simple numeric cap on concurrent plan jobs. It has no awareness of system resources (CPU, memory, disk) and cannot dynamically adjust scheduling based on load.

Proposed Solution

1. System resource monitoring: Check CPU load, available memory, and disk space before launching jobs
2. Dynamic throttling: Reduce concurrent jobs when system is under pressure, increase when resources free up
3. Per-job resource hints: Allow jobs to declare resource needs (e.g., resources: { memory: "high" }) for better scheduling
4. Backpressure: If system resources are critically low, pause new job launches and notify the user
5. Metrics exposure: Surface resource usage in mc_overview

Priority

P3 — long-term improvement. Current maxParallel works well enough for most workloads. This becomes important when running many heavy jobs (e.g., full builds, large test suites) on resource-constrained machines.