This is Windows Server 2016, is it? I think that was dynamic quorum at play. Because you lost the FSW you had a cluster with an even number of votes, which obviously isn't ideal because of the risk of split brain, so it would have removed the vote of one of the nodes until the FSW came back. What I'm a little hazy on is why it chose the server it did- not sure about that bit.
I don't think the cluster did actually fail over, by the way; it probably just adjusted the quorum as above, hence no AG failover.