optimize buffer protocol constructor for stl_bind std::vector when buffer is a simple array

[marc-chiesa]

When constructing a std::vector using a Python buffer protocol object, there is a simple optimization for step == 1 (i.e. a simple contiguous array) that can yield a significant performance boost, especially for large arrays.

Old code:

        auto vec = std::unique_ptr<Vector>(new Vector());
        vec->reserve((size_t) info.shape[0]);
        T *p = static_cast<T*>(info.ptr);
        ssize_t step = info.strides[0] / static_cast<ssize_t>(sizeof(T));
        T *end = p + info.shape[0] * step;
        for (; p != end; p += step)
            vec->push_back(*p);
        return vec.release();

New code:

        T *p = static_cast<T*>(info.ptr);
        ssize_t step = info.strides[0] / static_cast<ssize_t>(sizeof(T));
        T *end = p + info.shape[0] * step;
        if (step == 1) {
            auto vec = std::make_unique<Vector>(p, end);
            return vec.release();
        }
        else {
            auto vec = std::make_unique<Vector>();
            vec->reserve((size_t) info.shape[0]);
            for (; p != end; p += step)
                vec->push_back(*p);
            return vec.release();
        }

[bstaletic]

This was resolved in #2298. Thanks @marc-chiesa