memleak is detected during glusterd peer handshake

[chen1195585098]

asan reports:

Direct leak of 4386 byte(s) in 34 object(s) allocated from:
    #0 0x7ff3b72efbb8 in __interceptor_malloc (/lib64/libasan.so.5+0xefbb8)
    #1 0x7ff3b6902420 in __gf_malloc (/lib64/libglusterfs.so.0+0x102420)
    #2 0x7ff3b6876e18 in dict_allocate_and_serialize (/lib64/libglusterfs.so.0+0x76e18)
    #3 0x7ff3aa5ec701 in glusterd_mgmt_handshake glusterd-handshake.c:2212
    #4 0x7ff3aa5ee2cc in __glusterd_peer_dump_version_cbk glusterd-handshake.c:2388
    #5 0x7ff3aa5b58ca in glusterd_big_locked_cbk glusterd-rpc-ops.c:217
    #6 0x7ff3b642ffcb in rpc_clnt_handle_reply rpc-clnt.c:780
    #7 0x7ff3b64307e9 in rpc_clnt_notify rpc-clnt.c:957
    #8 0x7ff3b6425c2c in rpc_transport_notify (/lib64/libgfrpc.so.0+0x25c2c)
    #9 0x7ff3a880d874 in socket_event_poll_in_async socket.c:2531
    #10 0x7ff3a882828e in socket_event_poll_in ../../../../libglusterfs/src/glusterfs/async.h:189
    #11 0x7ff3a882828e in socket_event_handler socket.c:2963
    #12 0x7ff3a882828e in socket_event_handler socket.c:2883
    #13 0x7ff3b69af776 in event_dispatch_epoll_handler event-epoll.c:640
    #14 0x7ff3b69af776 in event_dispatch_epoll_worker event-epoll.c:751
    #15 0x7ff3b48082fe in start_thread pthread_create.c:479
    #16 0x7ff3b3e39dd2 in __GI___clone (/lib64/libc.so.6+0x39dd2)

By the way, size in mem_acct data struct, which indicates total size allocated for data-type, is not updated correctly.
size will increase when GF_MALLOC is called, however, it never get a decreasement in GF_FREE.
Should fix it here, otherwise the wrong size may mislead mempool memleak anlysis.