csum: Inline csum_add16(), csum_add32().

These two functions are really simple, to the point that inlining them
yields less code than calling them: with GCC 4.9 on i386, this change
reduces the code generated for packet_csum_pseudoheader() from 112 bytes
to 100 bytes.

I didn't measure whether there is a performance improvement.

Signed-off-by: Ben Pfaff <blp@ovn.org>
Acked-by: Thadeu Lima de Souza Cascardo <cascardo@redhat.com>