Genesis test failure: extra honest header download

[nfrisby]

$ git lg -1
99e49deff - (HEAD, origin/main, origin/HEAD) Add version `0.5` and its publishing date to the `utxo-db-api` doc (#1438) (21 hours ago) <Joris Dral>
$ cabal run ouroboros-consensus-diffusion:test:consensus-test -- --quickcheck-tests=1 --quickcheck-replay='(SMGen 5993563275997594115 6875817751050742035,0)' -p '/With some adversaries.honest peers do their own thing/' --quickcheck-shrinks=10000000

The test fails because the same header is downloaded from two honest peers, which CSJ is expected to avoid in the "idealized" scenario of these tests.

It wasn't obviously clear to me why it was happening in this case. But I've learned some things in in the other triage cases, and I have more suspicions now. This Issue is to return to this.

(This is not a blocker for the initial opt-in release of Genesis---it merely indicates the CSJ optimization is not a 100% optimal, which is harmless in the short-term.)