Scheduled runs route to dedicated node pools

Scheduled runs create predictable hourly spikes that compete with on-demand runs for node capacity. When a burst of scheduled runs lands at the top of the hour, it can saturate shared pool resources, causing slower cold starts for users waiting on on-demand results.

This PR introduces soft scheduling affinity rules that route scheduled runs to a dedicated pool while on-demand runs avoid it. The affinity is a preference, not a requirement — runs fall back gracefully if the target pool is out of capacity.

The supervisor checks the run's rootTriggerSource annotation at dequeue time. When it equals "schedule", the workload manager applies soft node affinity preferences toward the scheduled runs pool. Non-scheduled runs apply soft anti-affinity to avoid that pool.

In the supervisor app's Kubernetes workload manager, new environment variables control the behavior: KUBERNETES_SCHEDULE_AFFINITY_ENABLED toggles the feature, while KUBERNETES_SCHEDULE_AFFINITY_POOL_LABEL_VALUE and KUBERNETES_SCHEDULE_AFFINITY_WEIGHT configure which pool to target and how strongly. This follows the same pattern as existing large machine affinity, which was also refactored for consistency.

This is part of an ongoing effort to improve resource isolation and reduce cold start latency for on-demand workloads.