Use function pointers for runtime dispatching

Rather than branching each time, consider doing something like this:

int (*run_container_cardinality)(const run_container_t *run) = run_container_cardinality_dispatch;

int run_container_cardinality_dispatch(const run_container_t *run) {
    if (croaring_avx2()) {
        run_container_cardinality = _avx2_run_container_cardinality;
    } else {
        run_container_cardinality = _scalar_run_container_cardinality;
    }
    return run_container_cardinality(run);
}

This way after the first call all calls will be direct. It may give a tiny performance gain.

Start by tracing run_container_cardinality_dispatch, run_container_cardinality, croaring_avx2, and the scalar and AVX2 implementations mentioned in the issue. Verify that runtime dispatch preserves existing behavior, then measure whether subsequent calls gain the proposed performance improvement.