Describir: Deep Learning Scheduling on a Field-Programmable Gate Array Cluster Using Configurable Deep Learning Accelerators