Learning Conditionally Calibrated Equations of State for Direct Fired sCO2 Cycles with Deep Neural Networks