Make GhostVertexRemover easy to use

Nobody has claimed this yet.

JanusGraph codebase contains a `GhostVertexRemover` that implements `ScanJob` interface, which could be used via:

```java
JanusGraphManagement.IndexJobFuture ghostRemover =
graph.getBackend().buildEdgeScanJob()
.setJob(new GhostVertexRemover(graph))
.execute();
```

The above way, however, runs on a single instance, and thus its capacity is limited despite its use of multi-threading. It is unrealistic to use it to remove all ghost vertices on a massively large graph. We should make it easy to use in both 1) single machine 2) Hadoop system. We should, of course, also add the usage to the documentation

A good example is`IndexRemoveJob`, which also implements the `ScanJob` interface. It can be used via `ManagementSystem` (single machine) and `MapReduceIndexManagement` (map-reduce).

Start by comparing GhostVertexRemover with IndexRemoveJob, especially how ManagementSystem and MapReduceIndexManagement expose ScanJob implementations. Trace the existing single-machine execution path and the Hadoop example, then define the usage paths for both environments and document them. Done means GhostVertexRemover is usable on a single machine and through Hadoop, with documentation covering both.

Written by the indexing model from the issue text.