Introduce a switch for reverting twofold draw visits

[Naphthalin]

When #1375 was merged, we decided to revert all visits to twofold draws if they were made into non terminals to keep the tree visit consistent. However, this introduced two problematic interactions:

1. In rare cases it's possible that the repetition line was the only non-losing line for one side, and reverting all visits to that line specifically will make first move of that line look much worse (or better) than it actually is.
2. In fast time controls, reverting the visits will lead to other root moves having more visits, therefore being selected accidentally. This can by made even more likely by an interaction with 1, and happened in Evaluation Blindness to forced 3-fold repetition #1466

This PR introduces a switch whether to revert twofold visits when the twofold draw node is made non terminal; default behaviour is --revert-two-fold-visits=false for testing purposes, =true is equivalent to the current behaviour.

[Naphthalin]

from Discord, thx to trre:

decided to take a stab at it since it seems like MakeNotTerminal(terminal_visits) was doubling the visits the node should have. changed it to MakeNotTerminal(0) and kept the if (!revert_two_fold_visits) break; in there so nothing is reverted

[trre123]

With changes listed above, this test was run at 6s+0.1s

   # PLAYER        :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 lc0-pr1513    :     2.1   10.6   503.0    1000    50      65  133  740  127    74
   2 lc0-master    :     0.0   ----   497.0    1000    50     ---  127  740  133    74