Change subtrace_len computation algo in FoldFSM

[mikevoronov]

Previously, to handle scripts like this

(fold
   (par
      (next
       ...

there was a temporary solution based on additional field called total_subtrace_len. This solution is needed because of Par and Fold decribe their subtree lens differently with respect to next instruction: Par includes states that next brings, while Fold not.

The new algo changes this scheme and does a convolution of a Fold subtrace_len. Let's it on examples, imagine a fold on stream with 3 elements that have the same generation, in this case convolution will look like this:

[1, 1] [2, 2] [3, 3] => [6, 1] [5, 3] [3, 6]
  g0     g0     g0

here a number before comma represents count of elements before next, and after the comma - after

For fold with 5 elements of two generations:

[1, 1] [2, 2] [3, 3] [4, 4] [5, 5] [1, 1] => [6, 1] [5, 3] [3, 6] [9, 4] [5, 9] [1, 1]
  g0     g0     g0     g1     g1     g2

This algo allows us to get rid of total_subtrace_len and all code that supports it.

Resolves #135.