How does it handle the mismatch between jsonb and bson? In particular:
* more scalar types, including blobs. One hack I could think of is using ISO 8859-1 encoding instead of UTF-8, which can act as a pseudo-blob.
* preserved field order
The order of fields is maintained by adding a special field with keys in the original order. For example, `{z: 1, a: {y:2, x:3}}` is stored as `{"a":{"x":3,"y":2,"$k":["y","x"]},"z":1,"$k":["z","a"]}`.
Additional types are stored as objects with special fields too. For example, binary values are stored as `{"$b": "<base 64 string>", "s": <subtype number>}`. The full mapping is there: https://github.com/FerretDB/FerretDB/blob/b7e8240607e043a858...