Fix FILL/COMPLETION ring element sizes in the AF_XDP page

[NAME-ASHWANIYADAV]

Fixes #239

The XSK setup example sized all four ring mmaps with sizeof(struct xdp_desc), but the FILL and COMPLETION rings are arrays of plain _u64 UMEM addresses, not descriptors. The same mistake appeared in the ring-access line right below the mmap call, and in the ring-buffer concept section further down the page.

Verified against libxdp and the kernel

• libxdp's xsk_umem_create_with_fd maps the fill ring with off.fr.desc + fill_size * sizeof(_u64) (and the completion ring likewise), while the RX ring uses off.rx.desc + rx_size * sizeof(struct xdp_desc) - exactly the split the reporter pointed at (xsk.c#L258, xsk.c#L1167)
• Kernel-side, the two ring layouts are xdp_rxtx_ring { …; struct xdp_desc desc[] } vs xdp_umem_ring { …; u64 desc[] } (net/xdp/xsk_queue.h)

Changes

• The mmap length now uses the four-way expansion sizeof({struct xdp_desc,struct xdp_desc,__u64,__u64}), matching the page's existing {rx,tx,fill,completion} shorthand one-to-one
• The ring-access line is split so RX/TX rings are typed struct xdp_desc * and FILL/COMPLETION rings __u64 * (the old single line also used invalid array syntax)
• A short note under the example explains the element-size difference
• The concept section's "array of descriptors (struct xdp_desc)" now names both entry types

[NAME-ASHWANIYADAV]

@dylandreimerink will appreciate a review !!