Missing features and traversal strategies

[tixxit]

The goal of this PR is really to allow new ways to deal with missing features when evaluating a tree. Some changes to support this:

• All original predicates return true if a feature value is missing,
• A new predicate (IsPresent) was added which returns false if a feature is missing,
• New tree traversal strategies:
  • FirstMatch - returns first node whose predicate is true,
  • MaxWeightedMatch - returns node whose predicate is true with the most weight (annotation),
  • ProbabilisticWeightedMatch - returns random node whose predicate is true, but the probability of being chosen is weighted (annotation).

[avibryant]

LGTM modulo comments.