data are broken out into a new struct span_arrays which is allocated per-context (to avoid huge stack allocations - a problem on Windows). This lets us use span.redStep instead of span->redStep (for example) to hopefully get slightly better performance in the triangle functions.