When decoding some strings containing unpaired surrogates, UTF-16 decoder produces wrong number of \ufffd runes. Some examples:

• On string \xdc\x00\xdc\x00: expected result \ufffd\ufffd (two copies of \ufffd), actual result \ufffd.
• On string \xd8\x00\x00: expected result \ufffd, actual result \ufffd\ufffd.

The expected results are derived from a WhatWG spec.

Also, the name of internal function isHighSurrogate is misleading: it actually checks whether the argument is a low surrogate.

Code to reproduce:

package main

import (
	"golang.org/x/text/encoding/unicode"
	"fmt"
)

func main() {
	res, err := unicode.UTF16(unicode.BigEndian, unicode.IgnoreBOM).NewDecoder().String("\xd8\x00 ")
	fmt.Println(res, err)
}