Linear Regression

Univariate linear regression via gradient descent, with visualization and error metrics. No ML libraries — the goal was to understand the mechanics directly, not to call a fitting function.