Refactor `TensorNetwork` to class-based OOP organization

[mofeing]

This PR starts the process of rewriting current TensorNetwork type hierarchy on more OOP-like classes. It doesn't do all the things we would like from it yet, but we need some other changes and this PR merges the minimum changes in order to proceed with them.

The main change is that we get rid of the "field-inheritance hack using NamedTuple", which although interesting, it doesn't scale well. It also redefines what an Ansatz is and turns Boundary and Plug into traits.

Things yet to do after this PR

• [ ] Refactor TensorNetwork internals to encode the dual graph of the Tensor Network #104
• [ ] Implement "multilayer-graph"-like functionality in QuantumTensorNetwork #105
• [ ] Design Ansatz subtypes as special cases of graph layers #106
  • Graph layers should encode the entanglement connectivity pattern (i.e. virtual indices)
  • The Ansatz IS the graph layer (e.g. a MPS and a MPO have the same "ansatz" and graph layer)
• [x] Consider deprecating Infinite plug trait #107
  • Infinite graphs are difficult to manage
  • Many "infinite" algorithms can actually be implemented by using "open" tensor networks
• [ ] Discuss composition over field inheritance for the TensorNetwork hierarchy
  • Currently we are doing field inheritance with Classes.@class
• [ ] Consider a Classes.jl fork or build on top of other packages #108

[codecov[bot]]

Codecov Report

All modified lines are covered by tests :white_check_mark:

:exclamation: No coverage uploaded for pull request base (develop@be2e2e9). Click here to learn what that means.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## develop #103 +/- ## ========================================= Coverage ? 0.23% ========================================= Files ? 14 Lines ? 869 Branches ? 0 ========================================= Hits ? 2 Misses ? 867 Partials ? 0 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.