Cross-Lingual Alignment Without Joint Training: Do Monolingual Language Models Converge on Universal Representations?