Statistical Embeddings for Similarity, Retrieval, and Interpretable Alignment of Numeric Tabular Datasets — ThinkLLM