fix: fail e2e scenarios when a required node dies - #416

I think it is not completely fixed yet.

Some of the predicates are still partial. For example, stake_and_checkpoint and stake_and_join_with_outdated_ckpt still skip node0 in the final progress loop, and withdraw_and_exit / sync_from_genesis still check progress only for 0..(NUM_NODES - 1). That may be intentional for stopped/withdrawn nodes, but the scenario still does not explicitly assert the expected shutdown/removal outcome for every omitted participant.

There are also still join paths that only print thread join failures and then continue to “All nodes shut down cleanly” / exit(0), for example in observer and the checkpoint scenarios. testnet.rs similarly logs inner node errors from try_join_all but does not fail the process.

Each scenario should define the required participant set and either assert progress for each required node or assert the expected clean shutdown/removal for intentionally stopped/withdrawn nodes, and join failures should become nonzero scenario failures.

Update(b4507b1):

• Fail e2e scenarios on unexpected node-handle exit or join failure

Thanks.

I think there is one remaining similar path in protocol_params.rs: the node wrapper still treats _ = node_handle.fuse() as just println!("Node {} handle completed", x). Since there is no expected withdrawn/stopped participant in that scenario, an early handle completion should probably be treated the same way as the other scenarios and fail the scenario.

So I think the update fixes my comment for the touched paths, but I would still patch protocol_params.rs before closing #318 completely.

Update(018839f):

• Fail e2e scenarios on remaining unguarded node-exit paths

This looks good to me.