[chandlerhuff]

Summary

Makes router and executor placement resource-aware so VM starts are limited by aggregate CPU, memory, and VM slots instead of VM count alone. This lets mixed-size jobs share a host when they fit while keeping oversized concurrent jobs queued.

Changes

• Added shared resource-capacity accounting and backward-compatible executor default resource reporting.
• Updated the router to enforce per-job caps plus live aggregate capacity and reserve CPU/memory immediately after dispatch.
• Updated executor webhook and management dispatch paths to reject or queue work that cannot fit, while allowing smaller pending jobs to backfill available capacity.
• Added capacity-specific logging, OpenAPI/README documentation, and 12 focused tests across common, router, and executor packages.

Testing

• swift test --package-path Packages/TartCommon
• swift test --package-path Packages/RouterApp
• swift test --package-path Packages/ExecutorApp
• swiftlint lint --quiet --config .swiftlint.yml
• Parsed both OpenAPI YAML resources successfully.
• Note: the root swift build was blocked while downloading the external SwiftLint artifact; the changed packages compiled and passed independently.

---

Ticket: NO-TICKET (no Jira ticket)