It would be useful for $bsonSize & related operators to work with over-large documents. Then change streams could do stuff like:

cs = db.watch(
    [
        {$set: {
            fullDocument: { $cond: [
                {$gt: [{$bsonSize: "$$ROOT"}, 16 << 20]},
                "truncated",
                "$fullDocument",
            ]},
        }},
    ],
    {fullDocument: "updateLookup"},
);

const longString = "x".repeat(10 << 20);

db.foo.insert({_id: 1});
db.foo.updateOne({_id: 1}, {$set: {long: longString}});

The change stream’s `update` change event will exceed 16 MiB because it contains the update description—including the full 10-MiB string—as well as the `fullDocument`, which also includes the string. The aggregation tries to truncate `fullDocument` for such oversized events. The attempt fails, though, because $bsonSize itself can’t report the size of oversized documents.