avoid size_t overflow in zip_central_dir_move middle-run check

[jmestwa-coder]

middle-run guard in zip_central_dir_move uses a size_t multiply to mean both sides have survivors:

• l_size * r_size wraps to 0 on a 32-bit size_t build when both are a multiple of 64 KiB (reachable via ~64 KiB entry names)
• the branch is skipped, so surviving records after the deleted run are not shifted over the gap and their central-dir offsets not rebased, while m_central_dir.m_size is still lowered
• the finalized archive's central directory is left desynced from its offset table

test each size against zero directly; behavior is identical on 64-bit.