Per-request bounds so a single slow/hung provider can't stall the whole request indefinitely once fetching happens concurrently.