HIgh Block Production Time

[PropzSaladaz]

Description

There seems to be a bug somewhere in consensus logic that is slowing down block production.
Data supporting this issue:

1. Block production time was decreased from ~11s to 3s after all BITE perf. improvements from Consensus tests have high TLWT #946 . Varying number of txs from 1-300 only skews block production time by 200ms over 3s. So the problem is not related to BITE processing anymore
2. Plots show relatively large sections of unused CPU that appear consistently, but have no predictable pattern (see below - first rectangle represents around 1s, and the second rectangle around 500ms - way too much time for 100ms communication rounds)

3. Using only 1 tx leads to ~3s block time.

The 3s latency was measured with 100ms simulated latency for each network message.