RFC 8259 section 7 requires special handling of surrogate pairs like "\uD834\uDd1e":

To escape an extended character that is not in the Basic Multilingual
Plane, the character is represented as a 12-character sequence,
encoding the UTF-16 surrogate pair. So, for example, a string
containing only the G clef character (U+1D11E) may be represented as
"\uD834\uDD1E".

`UnmarshalExtJSON` does not properly decode surrogate pairs. Instead it converts each to a Unicode replacement character.

Demo program:

package main

import (
	"encoding/hex"
	"fmt"

	"go.mongodb.org/mongo-driver/bson"
)

func main() {
	str := `{"a":"\uD834\uDd1e"}`
	doc := bson.D{{"a", "\U0001D11E"}}

	var buf bson.Raw
	err := bson.UnmarshalExtJSON([]byte(str), true, &buf)
	if err != nil {
		panic(err)
	}
	fmt.Println("Unmarshaled from JSON: " + hex.EncodeToString(buf))

	doc2, err := bson.Marshal(doc)
	fmt.Println("Marshaled from bson.D: " + hex.EncodeToString(doc2))
}

Output:

Unmarshaled from JSON: 1300000002610007000000efbfbdefbfbd0000
Marshaled from bson.D: 1100000002610005000000f09d849e0000

Treatment if ill-formed surrogate pairs (e.g. only one) is often implementation defined. You can find cases to consider in this corpus: https://github.com/nst/JSONTestSuite