Kuroiwa's Biased Exploration code base ported to modern fast downward

[guicho271828]

This PR adds three OPEN list implementations (Kuroiwa and Beck, ICAPS2022, https://github.com/Kurorororo/biased-exploration) ported for the current fast downward. I am not the original author of the source code. I reproduced the results for Softmin (the most performant) and it is quite impressive.

• softmin : Open list that randomly selects h with a probability that is a softmax over different h values. Works with a single heuristic function.
• softmin_heap : same, but uses a heap.
• softmin_type_based : Similar to type_based, it works with multiple evaluators. Instead of selecting buckets uniform randomly, uses the softmin criteria.
• linear_weighted, linear_weighted_heap, linear_weighted_type_based : Open list that randomly selects h with a probability that is linearly interpolated between max/min h, and its variants.
• nth, nth_type_based : Open list that randomly selects from the n th best element.

I am open to adding softmin and softmin_type_based variants only, removing the less-performant variants. softmin_heap may be useful depending on the action cost.

Another opportunity might be to remove existing type-based open list, given that softmin_type_based is better theoretically & empirically.