[Bug fix] Support for to_string calls with big Uint8Arrays

[abenmrad]

It seems that, despite not being defined the ECMAScript spec, JavaScript runtimes tend to impose a limit on how many arguments a function can receive. And since the current to_string implementation in the libsodium wrapper uses a Function.call paradigm, you tend to hit that wall at some point.

Here is a fix (among other possible ones) :

/*
The main idea is to split the buffer into a smaller buffers,
call to_string on each one of them, and concatenate the results
*/
var toStringChunkSize = 32767;

function to_string(bytes) {
        if (typeof TextDecoder === "function") {
            return new TextDecoder("utf-8", {fatal: true}).decode(bytes);
        }

    var numChunks = Math.ceil(bytes.length / toStringChunkSize);
    if (numChunks > 1){
        var totalString = '';
        for (var i = 0; i < numChunks; i++){
            totalString += to_string(Array.prototype.slice.call(bytes, i * toStringChunkSize, (i + 1) * toStringChunkSize));
        }
        return totalString;
    }

    try {
        return decodeURIComponent(escape(String.fromCharCode.apply(null, bytes)));
    }
    catch (_) {
        throw new TypeError("The encoded data was not valid.");
    }
}

PS: Sorry for not submitting this as a PR.

[jedisct1]

Unfortunately, the bytes cannot be split at arbitrary positions, because the split could then happen in the middle of a UTF-8 sequence :(