Build `confusion_matrix_accuracy(tp, tn, fp, fn)` — the foundational classification metric: - Formula: `(tp + tn) / (tp + tn + fp + fn)`. - Return float in `[0.0, 1.0]` rounded to 4 decimals. - All four zeros → return `0.0` (avoid division by zero — empty test set). - Validate non-negative; treat any negative input → return `0.0` (defensive). Accuracy is the right metric only when classes are roughly balanced — but every interviewer wants to see you implement it cleanly first.