The metaphonic diphthongisation phenomenon is said to have occurred between the 6th and 8th century. But it must have happened before the loss of intervocalic "v", though I have only one example to support my claim and i'm not sure if it's indeed proof of it. nivem/*nivis -> *neave(s) -> nea. I can't find any info on this now, but I remember reading of a vulgar latin place name in the Balkan peninsula mentioned in a 6th or 7th century greek document that presented the loss of intervocalic "v". this is my attempt to recall the word: "ketate" ( <- civitatis ). This shows that "v" was already disappearing in Balkan proto-romance. Would this imply that the so called "umlaut" ( diphthongisation ) phenomenon happened even earlier? Well, there is also the famous "Torna, torna, fratre" episode dated to 587 that doesn't show any sign of diphthongisation in the word "torna" ( expected: toarna ).
the alternative form "neauă" could also be explained from the accusative "nivem". According to wikipedia there was an intermediary stage before the loss of intervocalic "v" where v -> w. Could this stage have been preserved in "neauă" but not in "nea" ( that could be a shortened form )?
I know that all of this is a gross oversimplification, misuse of linguistic tools, and possibly misinterpretation of historical linguistics.I am just an amateur with an interest in Romance languages. So my bad for that.