To get a bit more technical, I distinguish between mastering and exporting here.
Mastering, in my book, is the (somewhat) artistic process of applying EQs and compressors/expanders in order to level out the frequency spectrum.
Exporting is the final step where you convert the software's internal representation of the audio (usually
floating point these days).
In my interpretation this happened:
1) DT records and mixes the album
2) The mix gets sent to mastering guy, who masters it
3) The (floating-point) master gets exported down to 44.1kHz, 16-bit for CD production
Now, the hard clipping got introduced in step 3, the exporting. Floating-point audio has no volume limitations, but when converting it to 16-bit audio, it had to clip certain samples because their values were too high.
HDTracks took the *same* master from after step 2), the one in floating-point. Then, it lowered the value of the master a bit, and instead of converting to 44kHz, 16-bit, converted down to 96kHz/24-bit. The volume decrease meant that fewer samples clipped, but otherwise it is exactly the same source master.