I think it's been there from the beginning, but I'm not 100% sure. Realistically, Inf and -Inf are pretty much never useful in the real world, and having them makes -0.0 almost seem reasonable (since branch cuts almost always result in Inf/-Inf).
However before the standard was adopted it was decided to reduce the number of configurable modes and the single-value infinity was dropped.
The 2017 Gustafson paper linked in the article has only inf, no nar