Target Defense with Multiple Defenders and an Agile Attacker via Residual Policy Learning