On the Sample Complexity of Privately Learning Unbounded High-Dimensional Gaussians