Refactoring: Using tuples instead of dataclasses for Utility Analysis accumulators.

[dvadym]

This is done for performance reasons - it's much faster to work with tuples and serialized size of tuples is ~10 times smaller.

It turned out that the changes are pretty small and the resulting code structuring is nice:

1. Before:

   
   def create_accumulators():
   ...
   return Accumulator(...)

After

def create_accumulators(): ... return (...) # the same code, just Accumulator name dropped

2. Before

def compute_metrics(acc)

using acc.field1, acc.field2 etc

After

def compute_metrics(acc) field1, field2.. = acc

replace acc.field1 -> field1 ...

Also, it's introduced a base class for UtilityAnalysisCombiners.