Accepts unpaired surrogate codepoints in both encode and decode

[DavidBuchanan314]

import { encode, decode } from 'cborg'

const wtf8 = "😎".slice(1);
console.log(wtf8); // �

const encoded = encode(wtf8); // I expect an error here but there isn't one (or a U+FFFD substitution)
console.log(encoded); // Uint8Array(4) [ 99, 237, 184, 142 ] 

const decoded = decode(new Uint8Array([ 99, 237, 184, 142 ])); // I expect an error here although U+FFFD substitution is also reasonable
console.log(decoded); // ��� (this is in fact U+FFFD repeated 3x)

const roundtrip = encode(decoded);
console.log(roundtrip); // Uint8Array(10) [ 105, 239, 191, 189, 239, 191, 189, 239, 191, 189 ]

Encoding unpaired surrogate codepoints produces invalid utf8 byte sequences in the CBOR result.

Decoding the invalid utf8 produces strings with U+FFFD ("Replacement Character"), which is maybe reasonable, but other CBOR decoders may throw errors in the same situation.

[DavidBuchanan314]

I believe the bug only occurs for encode of strings of length <64, because the bug is in the optimized implementation only:

cborg/lib/byte-utils.js

Line 285 in 20f9e4f

function utf8ToBytes (str) {

(Buffer.from(string) does U+FFFD replacement - so the behaviour is inconsistent depending on input string length)