Storing the following values in BSONColumn opens object compression with a single interleaved stream for "x":

{"x": 1}
{"x": 2}
{"x": {"y": 2, "z": 3}}
{"x": {"y": 3, "z": 4}}

The {"y": 2, "z": 3} in the third and forth object are stored as uncompressed scalars in that single interleaved stream.

However, if the ordering of appends were reversed:

{"x": {"y": 3, "z": 4}}
{"x": {"y": 2, "z": 3}}
{"x": 2}
{"x": 1}

We would start with object compression with two streams and end it when the third object is encountered and restart object compression with a single stream for "x" as a scalar.

The behavior should be the same regardless of order. When going from scalar to object we should also re-start object compression.