vbucketmigrator with registered tap causes memcached to allocate 100GB of memory

1) start up 2 memcached processes with a single bucket
2) connect the memcacheds with vbucketmigrator using a registered tap name
3) check top to see virtual memory at 100gb
4) check mem_used in stats and see 100gb

This was only reproducible with vbucketmigrator, so we don't see it in the standard membase deployment.

commit id in ep-engine that it seems this was introduced: 2a9a5e96da62e21cf978a93b946f49c78adc8104

/opt/membase/bin/memcached -c 80000 -p 11211 -E /opt/membase/lib/memcached/ep.so -r -e "tap_idle_timeout=300;ht_size=12582917;dbname=/ebs/membase/default;min_data_age=0;queue_age_cap=900;max_size=13958643712;initfile=/tmp/init.sql;vb0=true" -u membase
/opt/membase/bin/memcached -c 80000 -p 11212 -E /opt/membase/lib/memcached/ep.so -r -e "tap_idle_timeout=300;ht_size=12582917;dbname=/ebs/membase/default;min_data_age=0;queue_age_cap=900;max_size=13958643712;initfile=/tmp/init.sql;vb0=true" -u membase

/opt/membase/bin/mbvbucketctl localhost:11212 set 0 replica

/opt/membase/bin/vbucketmigrator/vbucketmigrator -A -N replica -r -h localhost:11211 -d localhost:11212 -v -b 0