Possible LegacyDocument long content transcoding bug

Attached is a simple unit test.

A document content represents an incremented counter.

In tests, the document content (Long) is incremented 10 times from 0 to 10.
With even increments, the document representation is LegacyDocument. The content is cast to long, incremented, and the document is *replace*d
With odd increments, the document representation is LegacyDocument. The content is cast to long, incremented, and the document is upsert*ed as *JsonLongDocument

This replacement simplifies a business scenario and in reality should happened only once (to be followed with successive calls to bucket.counter

Something goes wrong during the 9 increment, the content is correctly read (8) and incremented (9).

However, during the 10 increment  - bucket.upsert(JsonLongDocument) - the value is reset before increment (testInc method simulate that).

Also, the value returned by bucket.get is not as expected and differ from the value returned  (testGet method simulate that).