Inconsistent results in equivalent queries

Version: 0.6.2

Query 1:
```
g.V().as('cccccccc').hasLabel('A').emit(__.loops().is(gte(1))).repeat(__.outE().inV()).times(2).emit(__.loops().is(gte(1))).repeat(__.outE().inV()).times(2).as('hketwcbf').hasLabel('B').bothE().dedup().by(__.path()).otherV().as('test').where(__.select('test').where(eq('cccccccc'))).select('hketwcbf').count()
```
Query 2:
```
g.V().as('cccccccc').hasLabel('A').bothE().dedup().by(__.path()).otherV().as('hketwcbf').hasLabel('B').emit(__.loops().is(gte(1))).repeat(__.inE().outV()).times(2).emit(__.loops().is(gte(1))).repeat(__.inE().outV()).times(2).as('test').where(__.select('test').where(eq('cccccccc'))).select('hketwcbf').count()
```

They are equivalent queries that have symmetric graph pattern. They would look like these in Cypher:
```
Q1: MATCH (cccccccc:A)-[]-(hketwcbf:B)<-[*1..2]-(ahdnqvvl)<-[*1..2]-(cccccccc:A) RETURN count(hketwcbf)
Q2: MATCH (cccccccc:A)-[*1..2]->(ahdnqvvl)-[*1..2]->(hketwcbf:B)-[]-(cccccccc:A) RETURN count(hketwcbf)
```

To reproduce:
```
g.addV('A').as('n1').property(single, 'name', 56).property(single, 'id', 2).property(single, 'undefined', 18).addV('A').as('n2').property(single, 'name', 61).property(single, 'id', 61).addV('A').as('n3').property(single, 'name', 49).property(single, 'id', 92).property(single, 'undefined', 16).addV('B').as('n4').property(single, 'name', 79).property(single, 'id', 47).addV('B').as('n5').property(single, 'name', 22).property(single, 'id', 17).property(single, 'undefined', 74).addV('A').as('n6').property(single, 'name', 99).property(single, 'id', 95).addV('B').as('n7').property(single, 'name', 29).property(single, 'id', 25).property(single, 'undefined', 1).addV('A').as('n8').property(single, 'name', 65).property(single, 'id', 62).addV('B').as('n9').property(single, 'name', 34).property(single, 'id', 83).property(single, 'undefined', 24).addV('B').as('n10').property(single, 'name', 91).property(single, 'id', 74).addE('x').from('n1').to('n8').addE('y').from('n2').to('n9').addE('x').from('n3').to('n3').addE('y').from('n4').to('n1').addE('x').from('n5').to('n5').addE('y').from('n6').to('n5').addE('x').from('n7').to('n6').addE('y').from('n8').to('n7').addE('x').from('n9').to('n5').addE('y').from('n10').to('n5').addE('x').from('n1').to('n5').addE('y').from('n2').to('n8').addE('x').from('n3').to('n2').addE('y').from('n4').to('n3').addE('x').from('n5').to('n6').addE('y').from('n6').to('n10').addE('x').from('n7').to('n7').addE('y').from('n8').to('n8').addE('x').from('n9').to('n9').addE('y').from('n10').to('n9').addE('x').from('n1').to('n8').addE('y').from('n2').to('n3').addE('x').from('n3').to('n1').addE('y').from('n4').to('n6').addE('x').from('n5').to('n9').addE('y').from('n6').to('n6').addE('x').from('n7').to('n5').addE('y').from('n8').to('n5').addE('x').from('n9').to('n10').addE('y').from('n10').to('n3').addE('x').from('n1').to('n10').addE('y').from('n2').to('n6').addE('x').from('n3').to('n2').addE('y').from('n4').to('n4').addE('x').from('n5').to('n3').addE('y').from('n6').to('n8').addE('x').from('n7').to('n4').addE('y').from('n8').to('n3').addE('x').from('n9').to('n8').addE('y').from('n10').to('n9').addE('x').from('n1').to('n9').addE('y').from('n2').to('n5').addE('x').from('n3').to('n7').addE('y').from('n4').to('n9').addE('x').from('n5').to('n5').addE('y').from('n6').to('n5').addE('x').from('n7').to('n7').addE('y').from('n8').to('n7').addE('x').from('n9').to('n9').addE('y').from('n10').to('n5').barrier().limit(0)
```

Run the supplied graph construction and both Gremlin queries against JanusGraph 0.6.2, then compare their counts and execution behavior. Use the stated symmetric graph pattern and Cypher equivalents to determine whether the results should match; done means the equivalent queries return consistent results.

Written by the indexing model from the issue text.