Implement Gini Impurity and information gain from scratch

Write pure NumPy functions to calculate Gini Impurity for a set of labels and evaluate the Information Gain for a proposed binary split on a continuous feature. Test it on a small subset of the data and verify it correctly identifies the best split point.

The issue names no file or test; start by locating the notebook or entry point that contains the decision-tree work and inspect the small data subset used there. Implement the pure NumPy calculations described in the issue, then verify the selected binary split point against the expected best split.

Written by the indexing model from the issue text.