DCTDecode is entirely legal for 1-component (greyscale) samples as you've noted. There is no problem there.
Remember, that applying a filter (and later decoding it) is entirely separate from the meaning of the data. So it's possible to use filters that just don't map correctly to the data. For example, greyscale DCTDecode could be used to encode samples which were, on decoding, to be treated as RGB. It wouldn't compress well, but the meaning isn't affected.
But DCTDecode for an Indexed colour space is very unusual. Don't look for special treatment, it would be enconding the image data, which would be offsets into the colour table. And of course DCTDecode is lossy, so decompressed data should be close, but is not always equal, to the original.
In most cases of Indexed colour, this would lead to random and useless changes to colours. But it is entirely possible to have colours which map into a linear and smooth set of shades of the same colour - such as 256 shades of pink. An image using such an Indexed colour space is a candidate for using DCTDecode.
In using filters, it is best to forget what you think you know about the data, and just follow the precise rules of the filter.