Implement Linear Regression Training

Implement a function `train_linear_regression(X, y, lr, epochs)` that trains a linear regression model from scratch using batch gradient descent. Requirements: - `X` is an `n x d` matrix of real-valued features. - `y` is a length-`n` vector of real-valued targets. - Initialize all weights and the bias to `0`. - On each epoch, compute predictions `y_hat = Xw + b`. - Use mean squared error as the loss. - Compute the gradients for `w` and `b`. - Update parameters with the given learning rate. - Return the final weight vector and bias. Do not call a library routine that directly fits a regression model. Basic matrix or array operations are allowed.