Function parallel_for_each 

pub fn parallel_for_each<I, F, T>(gate: RuntimeGate, tasks: I) -> Vec<Option<T>>
where
    I: IntoIterator<Item = F>,
    F: FnOnce() -> T + Send + 'static,
    T: Send + 'static,

Runs each closure in tasks on its own fiber and waits for all of them to finish before returning. Closures produce a T which is collected into the returned Vec in input order.

Panics inside individual tasks are caught via std::panic::catch_unwind and mapped to None in the output; the returned Vec contains all slots including None values for panicking tasks, preserving input order.

Uses a lock-free write path: each fiber writes directly into its own pre-allocated slot in an UnsafeCell<Vec<Option<T>>> using the slot index as the exclusive key — no Mutex contention between workers. The fan-in join barrier (dtact_await) provides the happens-before edge that makes the final read of all slots safe.

This is the workhorse used by parallel::solver and ffi::async_bridge to replace the std::thread::spawn pattern.