Adaptive posterior concentration rates for sparse high-dimensional linear regression with random design and unknown error variance