This is still an issue in beta5.
It has now been confirmed reproduceable between CHR instances in a lab environment, (see here and here), and the problem seems to be worse than I previously realised.
In my case, I can work around it with a firewall rule to block RA packets that lack prefix info, so that my default route doesn't get deleted, but it seems that in other cases some ISPs may not be sending any RAs with prefix info, in which case a default route never gets created.
This could result in an unexplained loss of outbound IPv6 connectivity after an upgrade from ≤7.20 to ≥7.21. (and already has done for one user), depending on how an ISP configures their RAs.