From Activation to Initialization: Scaling Insights for Optimizing Neural Fields