Training Sparse Mixture Of Experts Text Embedding Models