Every photo you upload is prepared before it is used: rotated upright, converted to a single format and compressed to a workable size. That preparation was discarding the information embedded in your file — the EXIF block written by your camera or phone, and any XMP entries such as author or credit.
Whether the information survived depended on how the particular file happened to compress, which is why the behaviour looked inconsistent from the outside: the same photo could keep its metadata one day and lose it the next, and files that needed no preparation at all kept theirs. In practice, most prepared photos lost it.
Photos are now handled the same way regardless of how they compress, and the embedded information is carried through to the file we store and that you download. Colour handling and the upright orientation are unchanged — the orientation is still applied to the image itself, so nothing is rotated twice.
Small previews used to display your photos in the interface remain without this information. They are display copies; the full file you download is the one that carries it.