Add support for HPA behavior configuration to control scaling stabilization in Langgraph helm charts #458
Description
Problem:
Currently, the HPA template in both api-server and queue doesn't support configuring the behavior field, which is essential for controlling scaling stabilization windows and policies. This causes rapid scaling up and down (thrashing) in production environments, especially for applications with long-running tasks with 5-10 minute graph executions.
Without behavior configuration, HPA scales too aggressively:
- Scales down immediately when metrics drop, potentially interrupting long-running tasks
- Causes unnecessary pod churn and instability
- Increases costs due to frequent scaling operations
- Creates poor user experience with connection drops