Discopy

* Fix bug for legacy mixed circuit eval (11d64)
* `.subs()` is fixed (see 45)
* Fix minor bugs (28cb4, 82cd5, a1910, b7235)

* Long range `Controlled` gates (see 59)
* Easy definition of pregroup diagrams and circuit diagrams using "chaining" (see 62)
* Easy permutation of monoidal diagrams using `.permute()` (now includes partial permutations)
* Better management of backend using `TensorBackend`, replaces global `Tensor.np` (58)
* Full `tensornetwork` support: `tensor.Diagram` and `Circuit`s (both mixed and pure) can be efficiently evaluated using Google's `tensornetwork` library. (see 60)

Support multi spider factory:

<img height="350" src="https://user-images.githubusercontent.com/13847804/145447552-c1f49b91-6081-40b3-9cc1-1dc56acc4688.png" />

Add support for pytorch backend with tensornetwork:
python
import tensornetwork as tn
import torch

from discopy import Tensor

tn.set_default_backend('pytorch')
torch.array = torch.as_tensor
Tensor.np = torch


then you can pass a contractor to eval.
python
diagram.eval(contractor=tn.contractors.auto)

Main Changes:

* all notebooks refreshed to reflect latest features
* conjugation of diagrams (vertical reflection)
* the ability to bend boxes within a diagram
* improve drawing of tensors and other spider diagrams

Main Changes:
* better drawing for pregroup diagrams
* new hypergraph module
* serialisation and deserialisation of diagrams
* conversion of rigid spiders into circuits