Google Drive API: Correct way to upload binary files via the Multipart API

I'm trying to upload a binary file to Google Drive via the multipart upload API v3.

Here's the hex representation of the content of the file:

FF FE

For some reason the above content gets encoded as UTF-8 (I assume) when I try to POST it, enclosed in a multipart payload:

--BOUNDARY
Content-Type: application/json

{"name": "F.ini"}

--BOUNDARY
Content-Type: application/octet-stream

ÿþ                <-- in the outbound request, this gets UTF-8 encoded
--BOUNDARY--

Hex representation of the file that ultimately gets stored on server side:

C3 BF C3 BE

The problem only occurs in the sending stage: if I check the length of the content read from the file I always get 2; regardless of whether I use FileReader#readAsBinaryString or FileReader#readAsArrayBuffer (producing a string with length 2, and an ArrayBuffer with byteLength 2, respectively).

Here's the minimal code that I'm using to generate the multipart payload:

file = picker.files[0];    // 'picker' is a file picker
reader = new FileReader();
reader.onload = function (e) {
    content = e.target.result;
    boundary = "BOUNDARY";
    meta = '{"name": "' + file.name + '"}';
    console.log(content.length);    // gives 2 as expected

    payload = [
        "--" + boundary, "Content-Type: application/json", "", meta, "", "--" + boundary,
        "Content-Type: application/octet-stream", "", content, "--" + boundary + "--"
    ].join("\r\n");
    console.log(payload.length);    // say this gives n

    xhr = new XMLHttpRequest();
    xhr.open("POST", "/", false);
    xhr.setRequestHeader("Content-Type", "multipart/related; boundary=" + boundary);
    xhr.send(payload);              // this produces a request with a 'Content-Length: n+2' header
                                    // (corresponding to the length increase due to UTF-8 encoding)
};
reader.readAsBinaryString(file);

My question is twofold:

• Is there a way to avoid this automatic UTF-8 encoding? (Probably not, because this answer implies that the UTF-8 encoding is part of the XHR spec.)
• If not, what is the correct way to "inform" the Drive API that my file content is UTF-8 encoded? I have tried these approaches, with no success:
  • appending ; charset=utf-8 or ; charset=UTF-8 to the binary part's Content-Type header
  • doing the same to the HTTP header on the parent request (Content-Type: multipart/related; boundary=blablabla, charset=utf-8; also tried replacing the comma with a semicolon)

I need the multipart API because AFAIU the "simple" API does not allow me to upload into a folder (it only accepts a filename as metadata, via the Slug HTTP header, whereas the JSON metadata object in the multipart case allows a parent folder ID to be specified as well). (Just thought of mentioning this because the "simple" API handles things correctly when I directly POST the File (from the picker) or ArrayBuffer (from FileReader#readAsArrayBuffer) as the XHR's payload.)

I do not want to utilize any third-party libraries because

• I want to keep things as light as possible, and
• keeping aside reinventing-the-wheel and best-practices stuff, anything that is accomplished by a third party library should be doable via plain JS as well (this is just a fun exercise).

For the sake of completeness I tried uploading the same file via the GDrive web interface, and it got uploaded just fine; however the web interface seems to base64-encode the payload, which I would rather like to avoid (as it unnecessarily bloats up the payload, esp. for larger payloads which is my eventual goal).