fix: surface unclamped demand in scale signal (#25)

[sylvesterdamgaard]

Summary

• Added demand field to each workload entry carrying the unclamped per-queue demand (from Phase A), separate from the fair-share-allocated target_workers
• Changed buildClusterSummary() to sum demand (not target_workers) when computing $requiredWorkers, so the scale-up condition in clusterScaleSignal() can fire when true demand exceeds cluster capacity
• No changes to FairShareAllocator, clusterScaleSignal, or recommendedHostCount — they already work correctly once fed the right input

Fixes #25

Test plan

• New test: 4 queues with demand summing to 30, capacity 10 → required_workers reports 30, scale_signal: scale_up, recommended_hosts > 2
• New test: demand fits within capacity → no false-positive scale_up, recommended_hosts == current_hosts
• New test: workload entries surface both demand and target_workers with correct values
• All existing test fixtures updated with demand field
• Full suite: 565 tests passed (1664 assertions)
• Larastan: no errors
• Pint: clean