If the recipient can prove he had the signed email before the signing private key was made public, eg, by contemporary third party notarization, the keys becoming public later don't help with disclaiming it at all.
The devil is right in the details... it is serving just as perfectly fine as a non-repudiation mechanism as it ever was, just constrained in time to a day or so as suggested by the article. Depending on what you said to whom that you wish you hadn't said, that can be all you need to get into trouble.
Publishing the privkey just makes it so in other circumstances where he might strongly wish that he had evidence he HAD said something, the dkim signature is devalued.
The handling rules standardized by DMARC are to ignore failing SPF when there is a valid DKIM signature, even if the domain doesn't use DMARC. Google, in particular, ignored SPF failures on DKIM signed messages last I checked.
Yes.
>Would this even truly stand up as believable for any scenario?
The post says this will most likely only work in a leak situation rather than a legal dispute. An example of where this might be useful: You send an email to a friend revealing something sensitive about yourself. A hacker hacks your friend's email and leaks it out. Without plausible deniability, everyone can verify cryptographically that you did in fact say that sensitive stuff (assuming your own email didn't get hacked). With this technique, people cannot verify whether or not you said it, the email might be a fraud and so people are less likely to believe the sensitive thing about you.