Fix callVariant with fusion in the same transcript

[zhuchcn]

As described in #342 that Arriba calls fusion in the same gene, which causes a problem of the graph data structure because we currently use transcript/gene ID as the subgraph ID to differentiate between the main graph and subgraph being inserted or fused. So in the PR, I redesigned data structure of how the relationship between subgraphs are handled. I use a tree-like structure (SubgraphTree), which is just simply a dict, that the keys are the subgraph ID, and the values are SubgraphLocation which has attributes of level, parent_id, and the location at the parent graph. I also switched to use a UUID for any subgraph ID just to avoid any collision.

Closes #341
Closes #342
Closes #343

[zhuchcn]

#347 is actually the same issue. In CVG, the variant graph is extended for another loop so it covers the variant peptides that span over the junction site. And previously the same transcript ID is used as subgraph_id, so it causes trouble when it tries to compare the relative position of two nodes over the junction site. It is now also updated to use the new SubgraphTree and use a unique subgraph_id to avoid this problem.

[lydiayliu]

ok i can confirm all the fusion cases were resolved. will merge now to play with the cpus setting and test the circ case