Function join_bounded 

pub async fn join_bounded<F, T>(futures: Vec<F>, limit: usize) -> Vec<T>
where
    F: Future<Output = T>,

Run a list of futures with bounded concurrency via Arc<Semaphore>.

Gate of record (rules_rust_paralelismo): each future acquires one permit with Semaphore::acquire_owned before polling body work; the permit is dropped (RAII) when the future completes. Internally composed with buffer_unordered so the stream polls at most limit futures, and the Semaphore is the admission control agents / tests observe.

Results are returned in completion order. Prefer this over unbounded join_all on collections of unknown size.

Cancel safety

Each future is polled independently; dropping the returned future cancels in-flight work at the next await point of those futures (permits return).

Observability

Host-local only: tracing::debug! of available_permits at start (no remote OTel — product law).

Gate pattern

Uses Semaphore::acquire (not acquire_owned) because callers pass borrowed CDP futures that are not 'static. Work stays on the same poller (buffer_unordered); permit is held for the future body and dropped via RAII. For tokio::spawn fan-out, call sites use acquire_owned + JoinSet instead (batch/crawl).