Fix the threshold in 3sf-mini

[konnov]

I believe that 3sf-mini has a rounding issue:

research/3sf-mini/consensus.py

Lines 96 to 97 in d225a67

	# If 2/3 voted for the same new valid hash to justify
	if count == (2 * state.config.num_validators) // 3:

For instance, if state.config.num_validators == 4, then the above condition evaluates to true on count == 2. This is due to the fact that distributed computing literature often refers to thresholds in rational numbers, e.g., $\frac{2n}{3}$, whereas integer division sometimes gives us a smaller threshold. As a result, if a faulty staker sends different blocks to different peers, the network may get partitioned. I have introduced such a proposer in a fork. Indeed, it keeps growing two and sometimes three chains. I've also noticed scenarios in the simulations, in which validators were forgetting finalized blocks:

This PR proposes a fix that avoids integer division:

        if 3 * count > 2 * state.config.num_validators:

With this fix in place, the simulation seems to work fine, even in the presence of a faulty proposer.

[konnov]

This is related to the post Simplifying the L1 by @vbuterin

[Arkeo35]

good work