Previously if we saw a mixture of int and double values in the same field in different documents in a collection we would infer the ExtendedJSON type for that field. This PR changes that behavior to infer double.

We could extend this to unify other compatible numeric types:

• int → long
• int → decimal
• double → decimal
• long → decimal

For now I've stuck with int → double because we have special handling an bson_to_json to accept a value of one type where the other is expected. We'll have to add similar special handling to support unifying the other numeric types.

MDB-128