implement serialization for Manellic tree

[dehann]

Description

Must solve kernel mixture and large nonparametric point count generic/JSONable serialization — e.g. 8192 points or larger; while equally efficient to replace mixtures.

Issues for HomotopyBelief are:

• (type stability problem) maybe only one kernel per belief such as a concentrated Gaussian,
  • eg one mean at root and then only tangents as you go down.
  • That's like building Huffman compression into the tree itself.
• Replacement for mixtures, such as Gaussian mixture models (GMMs)

Criteria

• think this is required to incorporate ManellicTree into IIF