Note that this makes pre-activation mempool behaviour slightly weird, as
noted in the test. In practice now that the upgrade has passed and is
checkpointed, only a new node during IBD can be in a preactivated state,
and a new node will anyway not request transactions before finishing IBD.
It may be worth adding tests of fresh-node IBD behaviour in a future
commit; similar behaviour occurs for any additive flag that is included
in mandatory flags.