It's funny; I wanted to upload a bunch of music from KeygenMusic.net into iTunes Match; but it wouldn't take them, because iTunes Match thinks that anything with a bitrate below, I think, 128kbps, is speech from a voice-recorder rather than professionally-mastered music.
In the end, I re-rendered all the music myself using a tracker program to FLAC, and from there to Q=256 constrained AAC. Each file was probably four times the size for no noticeable gain in sound quality, but at least they upload now. :)
That’s what I said I did, in the second paragraph. “Render” is the verb for going from a digital sequencer file to an analogue output (just like going from SVG to PNG is “rendering.”)
A digital encoding of Nyquist samples, though—meaning that you've got something which, at that point, is a lossless description of the equivalent analogue intensities that would be cut into a depth groove. (Provided that none of the data is above the Nyquist frequency, which is usually true-by-design in the kind of chiptune audio being discussed here.)
There isn't really any difference between holding onto the digital data after rendering it, and holding onto an (ideal, lossless) recording medium containing the analogue waveform you got by putting that digital data through a DAC. There's a bijection—you can always recover one from the other.
So why not think of PCM audio as an "analogue medium"—just one that requires a blob of digital logic to turn it into a signal on a wire, rather than a stylus?
If anyone else wants to do this: find one of the 7z files from pub.keygenmusic.org, grab OpenMPT [1] and export to WAV. It looks like ffmpeg can also link to libopenmpt but I haven't tried.
reply