app: Only perform spare check as of a certain cluster size.

[MathieuBordere]

run
https://github.com/canonical/jepsen.dqlite/actions/runs/7455383815/job/20284385625

assertion

 :assert {:valid? false,
          :count 1,
          :matches ({:node "n4",
                     :line "2024/01/09 01:29:31.730561 for jepsen: extra online spare"})},

log

2024/01/09 01:29:31.730561 for jepsen: extra online spare
2024/01/09 01:29:31.730580 changes={{3 3} map[{3297041220608546238 10.2.1.11:8081 voter}:0xc000428f20 {13470429068838499797 10.2.1.15:8081 spare}:0xc000428f10]} offlines=[] byFailureDomain=map[0:[{3297041220608546238 10.2.1.11:8081 voter} {13470429068838499797 10.2.1.15:8081 spare}]]

This looks like a false positive triggered by the following logic in go-dqlite:
https://github.com/canonical/go-dqlite/blob/4edab5e2dd71e25576b5095b2f3ebdea2664e1a0/app/roles.go#L135
The jepsen cluster doesn't have the minimum required size, because the member nemesis has removed a bunch of members.

The fix is to perform a specific check when the cluster doesn't have a certain minimum size.

[cole-miller]

Thanks, good point! I missed that wrinkle in the role management implementation. Would it be a good idea to tighten the code here to check that if the number of cluster members is less than the target number of voters, we have one voter + N spares?