Function dispatch_element_count_for_program 

pub fn dispatch_element_count_for_program(
    program: &Program,
    bindings: &[Binding],
) -> u32

Derive the dispatch element count from a binding plan and Program body.

Atomics and subgroup collectives execute over the logical input span even when they compact into a smaller output. Using the compact output length would suppress participating lanes before the atomic or collective runs.