Replicate the dynamic timeout issues in a knuu test

[evan-forbes]

When attempting to make the timeout commit dynamic, testground tests worked as intended but mocha failed to reach consensus.

The change can be found in #965

We should be able to replicate the failures noticed in an irl network in a test environment. This might involve randomizing the clocks for each instance, or adding more realistic values for latnecy.

Once we can reacreate the issue, we should be able to trace the round states and messages for all validators in the experiment. This should prove critical to debugging the underlying issues.

[evan-forbes]

blocked by celestiaorg/celestia-app#3488

[evan-forbes]

to update this, we were able to replicate this issue in knuu only when we restarted nodes. @cmwaters found that the start times differ after that and don't recover, which results in highly variable block times and missed rounds.