Use DataGraphs in ttn_svd and other changes

[emstoudenmire]

This PR contains mostly minor but helpful improvements to the ttn_svd function and its sub functions.

Here is a list of changes, from most to least significant:

• Use DataGraphs to hold the bond coefficients and sparse/symbolic vertex tensors returned from make_symbolic_ttn
• Reduce the number of arguments going into make_symbolic_ttn, svd_bond_coefs, and compress_ttn. (They weren't really sharing as much state as the code made it look, even compared to the current code following the last PR.)
• Significantly simplify svd_bond_coefs logic to just loop over all edges (it does a trivially parallel task of just forming and SVD'ing each edge coefficient matrix) rather than previous more complicated logic involving vertices and graph analysis.
• Removed other unneeded dictionaries.
• Use better graph analysis (vertex_path) to check which terms cross a given vertex. Could use steiner_tree but it seemed to have an issue with some disconnected components being left in the output.
• Remove coefficient_type function barrier in ttn_svd since now that function just calls out to other functions anyway.
• Simplify MatElem type keeping only code used for OpSum to TTN conversion. Will probably delete MatElem soon anyway in favor of SparseArray.