Add missing-value prediction policies for HoeffdingTree

[sekarminati]

This PR adds support for handling missing values during prediction in HoeffdingTree.

Previously, prediction followed MOA’s default behavior. With this change, users can choose how the tree should behave when a split attribute is missing.

What’s included:

• Add missing_value_policy to HoeffdingTree
• Support three policies:
  • default: use MOA’s default behavior
  • random: follow one randomly chosen child; if that path cannot produce usable votes, fall back to the current node
  • all: combine votes from all reachable children
• Add tree visualization
• Add tests for the new missing_value_policy parameter
• Add notebooks to demonstrate and experiment with the new behavior

[hmgomes]

Many thanks for the contribution! However, I am closing it for the time being