Ensure MG to have the same number of allreduce calls in mean_stddev for sparse matrix to avoid hanging

[lijinf2]

The hanging occurs when one GPU gets a sparse matrix of all zero values, while other GPUs get-zero values.