Scaling Laws for Precision in High-Dimensional Linear Regression