When one cites an article which appeared online-first in a journal of mathematics, but which is not in print and has no volume number yet, I have read that one can just use the DOI instead. But do you think that in this situation it is appropriate to add the link to the arXiv version anyway?
Here is an article about citing "advance online publication" in APA style.
Once an article is in the form of advanced online access at the publisher, you should generally cite this version, and not the ArXiv version. The online access is the final version of record. Page and volume numbers are cosmetic.
The exact form of the reference will depend on your citation style. But I agree that including the doi in the reference is helpful as this will not change. In contrast, the year of the reference may change.
Here is an example from the post above:
Muldoon, K., Towse, J., Simms, V., Perra, O., & Menzies, V. (2012). A longitudinal analysis of estimation, counting skills, and mathematical ability across the first school year. Developmental Psychology. Advance online publication. doi:10.1037/a0028240
So in general the reference really shows almost no difference to a standard reference. It's just that the volume and page numbers are missing. You might append some text to indicate that it is "advance online publication", but this is optional in many cases. And you should probably include the doi (either in doi:10.. form, or in http://dx.doi.org/10... form). The doi will be helpful especially where the paginated version comes out in a subsequent year.
Obviously, depending on your citation style, you'd need to adapt the above. But the basic concept is the same: Authors, year, title, journal, doi, optionally some indication that it is an advance online publication. Basically, follow your normal citation strategy but omit the volume and page details and include a doi.
In the citation that you use for your preprints you may use any citation style (the only criterion I follow there is that it should be as easy as possible to track down the references for anybody who reads the paper). For published versions, often the publisher decides how the references should look like. Sometimes they add dois (or other identifiers) themselves. Some may strip the link to the arXiv version and in that case you still may add the arXiv version as another entry and references it separately. Anyhow, if people search the paper by title, several search engines will bring up the arXiv version anyway.