limit() is not restrictive enough when combined with V()

- Version: 0.6.0 (and probably earlier)
- Storage Backend: any
- Mixed Index Backend: any

### Steps to reproduce
```groovy
g.addV().addV().addV().iterate()
```

### Expected Behavior (as produced by TinkerGraph)
```groovy
g.V().V().limit(1)
==>v[0]
g.V().V().limit(1).path()
==>path(v[0], v[0])
```
### Current Behavior:
```groovy
g.V().V().limit(1)
==>v[4152]
==>v[4152]
==>v[4152]
g.V().V().limit(1).path()
==>path(v[4152], v[4152])
==>path(v[4328], v[4152])
==>path(v[8424], v[4152])
```

Start by running the supplied g.V().V().limit(1) and path() reproductions against JanusGraph and comparing them with the stated TinkerGraph results. Trace the query handling for limit() combined with repeated V() steps; done means limit(1) returns one vertex and one corresponding path as shown.

Written by the indexing model from the issue text.