Resolve conflicts of MapNode with duplicate key caused by concurrent update to the same key with different type values other than primitive

Document state to JSON conversion introduced from #12 does not handle duplicate key as reported at [this comment](https://github.com/fthomas/crjdt/pull/12#issuecomment-270645358).

Duplicate key can be made when MapNode is updated with different type values other than primitive.

For example, concurrently updating to the same key "key" with object and string
```scala
val p1 = p0
.applyCmd(doc.downField("key") := `{}`)
.applyCmd(doc.downField("key").downField("key2") := "D")
val q1 = q0.applyCmd(doc.downField("key") := "C")
merge(p1, q1)
```

results in duplicate keys, RegT(StrK(key)) and MapT(StrK(key)).

```
MapNode(
Map(
MapT(DocK) -> MapNode(
Map(
RegT(StrK(key)) -> RegNode(Map(Id(2,q) -> Str(C))),
MapT(StrK(key)) -> MapNode(Map(RegT(StrK(key2)) -> RegNode(Map(Id(3,p) -> Str(D)))),Map(StrK(key2) -> Set(Id(3,p))))
),
Map(StrK(key) -> Set(Id(2,p), Id(3,p), Id(2,q)))
)
),
Map(DocK -> Set(Id(1,p), Id(2,p), Id(3,p), Id(2,q)))
)
```

Start with the document state to JSON conversion introduced from #12 and trace how MapNode handles concurrent updates during merge. Reproduce the Scala example with object and string values at the same key, then verify that the resulting state no longer contains both RegT(StrK(key)) and MapT(StrK(key)).

Written by the indexing model from the issue text.