You can use any available text as a cover. Of course depending on your algorithm design, some text might be better than others, therefore, there is not as far as I know any text document database that is commonly used as cover.
As mentioned above, typically, you can use any available text as a cover.
I found this dataset by Googling: https://github.com/mosser/steganography-dataset
This tool takes any text file and hides the message you want to embed in typos: http://mjethani.github.io/typo/
I think you could consider using text datasets from other domains such as NLP: https://lionbridge.ai/datasets/the-best-25-datasets-for-natural-language-processing/