Implement recursive decision tree builder

Build the core recursive function for the decision tree from scratch using NumPy. It should find the best feature and threshold to split on, and recursively build left and right branches until pure leaf nodes are reached. Show the training accuracy.

Start by locating the repository's Jupyter Notebook and the NumPy training code. Implement the recursive feature-and-threshold splitting described in the issue, stop at pure leaves, and show the resulting training accuracy.

Written by the indexing model from the issue text.