[cucumber] unignore leader failure tests and make them work

[sdbondi]

No description provided.

[Cifko]

Just a note why this isn't in progress.
The test works like this. We have committee of size n, then there is 1/n probability the one failed node will be hit in one block. So when we have transaction which is 3 blocks. There is a 1-(1-1/n)^3 probability we will actually trigger leader failure. That's quite high. But then we need a 3-chain to actually commit the transaction, which requires that we don't hit the failed leader in 3 consecutive blocks. Which is (1-1/n)^3. That's small.
In case of 4 nodes, there is ~58% chance that we trigger leader failure and ~42% chance that we form a 3 chain. That's flaky.