Integrate JunctionsTrees and Decompositions

[samuelsonric]

• Refactored all of JunctionsTrees.
• Added two constructors to Decompositions for creating a StrDecomp using graph elimination algorithms.

The two constructors have signatures

StrDecomp(
     graph::AbstractSymmetricGraph,
     ealg::Union{Order, EliminationAlgorithm}=DEFAULT_ELIMINATION_ALGORITHM,
     stype::SupernodeType=DEFAULT_SUPERNODE_TYPE)

StrDecomp(graph::AbstractSymmetricGraph, stree::SupernodeTree)

The type SupernodeTree encodes a supernodal elimination tree, an intermediate data structure used to construct a junction tree. It has a constructor

SupernodeTree(
     graph::AbstractSymmetricGraph,
     ealg::Union{Order, EliminationAlgorithm}=DEFAULT_ELIMINATION_ALGORITHM,
     stype::SupernodeType=DEFAULT_SUPERNODE_TYPE)

Supernodal elimination trees are cheaper to compute than full junction trees. Furthermore, they "know" the height and width of their corresponding junction trees. The following is a standard workflow in sparse matrix factorization.

1. Compute several supernodal elimination trees using different elimination algorithms.
2. Choose the tree with the smallest width to compute a junction tree.

---

Just updated the constructor to work on non-connected simple graphs by working on their connected components in parallel.