Towards Fine-Grained Multi-Dimensional Speech Understanding: Data Pipeline, Benchmark, and Model