fix(v2): use better seed for shard placement

[kolesnikovae]

The PR fixed a bad seed chosen for Fisher-Yates permutations forming the shard mapping in the new distribution algorithm.

I've noticed that in case of unavailability, the data distribution has pronounces skews and decided to test this case. I discovered that with low number of nodes and shards, the mapping we generate is very poor – the random choice is extremely unfortunate:

instance-0 [  35%][   6%] ###################################
instance-1 [  32%][  -3%] ################################
instance-2 [  32%][  -3%] ################################
-
instance-0 [   0%][ -99%]
instance-1 [  68%][ 105%] ####################################################################
instance-2 [  32%][  -3%] ################################
-
instance-0 [  35%][   6%] ###################################
instance-1 [   0%][ -99%]
instance-2 [  64%][  93%] ################################################################
-
instance-0 [  67%][ 102%] ###################################################################
instance-1 [  32%][  -3%] ################################
instance-2 [   0%][ -99%]

After some brute force (for 3-10 instances, where distribution skews are especially impactful) I found a better seed that ensures significantly better behaviour in case of node failures/unavailability:

instance-0 [  33%][   0%] #################################
instance-1 [  33%][   0%] #################################
instance-2 [  33%][   0%] #################################
-
instance-0 [   0%][ -99%]
instance-1 [  52%][  57%] ####################################################
instance-2 [  48%][  45%] ################################################
-
instance-0 [  50%][  51%] ##################################################
instance-1 [   0%][ -99%]
instance-2 [  49%][  48%] #################################################
-
instance-0 [  48%][  45%] ################################################
instance-1 [  51%][  54%] ###################################################
instance-2 [   0%][ -99%]

I'm not adding the scripts I used to find the seed, because I do not plan to maintain it: it's better to replace the randomized permutations with a knapsack solver or something similar that gives deterministic optimal results.

[simonswine]

Nice work! LGTM

[kolesnikovae]

Thank you Christian! This is a temporary workaround – I'm planing to revisit the approach we use to map shards to nodes