Improved recovery of the hemodynamic response in diffuse optical imaging using short optode separations and state-space modeling