Incorrect OR evaluation in JanusGraphStep

There's a noticeable difference in the evaluation of `or()` conditions together with vertex id. Depending on whether the initial vertex lookup is done with `g.V().has(T.id, ...)` or `g.V(...)`. When using
```
g.V().has(T.id, 1).or(__.has("prop", "A", __.has("prop", "B")))
```
the query optimizer (`JanusGraphStepStrategy`) folds all conditions into a `JanusGraphStep`. This seems to lead to an incorrect evaluation where the `or()` behaves like an `and()` - in `JanusGraphStep` all conditions are evaluated using `HasContainer.testAll()` method.

However, the query
```
g.V(1).or(__.has("prop", "A", __.has("prop", "B")))
```
produces the correct OR results, because folding is skipped in `JanusGraphStepStrategy`.

Start by reproducing the two queries and tracing the folding behavior in JanusGraphStepStrategy, then inspect JanusGraphStep and HasContainer.testAll(). Done means the g.V().has(T.id, ...) form preserves OR semantics like the g.V(...) form, without changing the correct behavior of the latter.

Written by the indexing model from the issue text.