The problem is memtx_tuple_new() says address of memtx_tuple, while
memtx_tuple_delete() says address of an actual tuple (part of memtx_tuple).
Different addresses in log may be confusing so the patch fixes the problem.

Closes #6634

NO_CHANGELOG=no visible changes
NO_DOC=no visible changes