Boosting Fast and High-Quality Speech Synthesis with Linear Diffusion