Small fix to the p2p test framework (mininode, etc) that I ran into when belatedly testing PR 7225.
I think it's better for solve() to recompute the hash before the first attempt so we're not using a cached hash, which if wrong, would almost certainly be a bug in the test.