Textnets

* Fix bug 36.
* Updates dependencies, including ``igraph`` and spaCy.

* This release removes Python 3.7 compatibility. You now need to use Python 3.8, 3.9 or 3.10 to use `textnets`.
* Fixes a bug in the HTML representation of the top-level module.
* Updates dependencies, including spaCy.

* Fixes 35 (invisible edges when scaling by weight).
* Updates some dependencies.

- Adds abilitiy to save and load an instance of `Corpus`, `Textnet` and `params` to and from file using `Corpus.save`, `tn.load_corpus`,
`Textnet.save`, `tn.load_textnet`, `tn.params.save` and `tn.params.load`. The same file can be used for all three kinds of objects, so all relevant data for a project can be saved in one file.
- Some further optimization of backbone extraction.
- Adds bipartite centrality measures (HITS, CoHITS and BiRank) and a bipartite clustering coefficient.
- Improved testing and type hints.
- Expanded documentation with advanced topics, including the new save/load feature and interacting with other libraries for network analysis and machine learning. Docs now use the PyData theme.
- Improvements to visualization. When plotting, nodes and edges can now be scaled by any attribute.
- **Breaking change:** Term weighing now happens in the `corpus` submodule, so the `sublinear` argument has to be passed to the methods for term extraction (`tokenized`, `noun_phrases` and `ngrams`). This change will make it easier to add additional term extraction and weighing options.
- Adds `tn.init_seed()` utility to quickly initialize pseudorandom number generator.
- Adds Python 3.10 compatibility.
- Updates dependencies, including `igraph` with some relevant upstream changes contributed by yours truly, as well as spaCy.

* Adds `params` as a container for global parameters. This makes it possible to fix the random seed and to change the resolution parameter for the community detection algorithm, among others. If the parameter `autodownload` is set to true, **textnets** will attempt to download all required spaCy language models automatically.
* Added HTML representation for the root module that displays versions of key dependencies.
* Added back string representations of `Corpus` and `TextnetBase`-derived classes.
* Adds a `Corpus.from_dict` method.
* `Corpus` now exposes the ``lang`` attribute, so the corpus language can be set after initialization of a class instance.
* The bipartite layout optionally used by `Textnet.plot` is now horizontal, so node types are arranged in columns rather than rows. That way node labels are less likely to overlap.
* Adds ``label_nodes`` argument to the `Textnet.plot` method to label both types of nodes. Defaults to ``False``.
* Adds ``node_opacity`` and ``edge_opacity`` arguments for `Textnet.plot`.
* Makes polygons marking clusters more visually appealing by adding opacity.
* Probably fixes [a bug](https://github.com/jboynyc/textnets/issues/30) that would occasionally result in an exception being raised during plotting (`IndexError: color index too large`).
* When initializing an instance of the `Textnet` class, you can now optionally pass the argument `connected=True`, in which case only the largest component of the underlying network is kept. When creating a one-mode projection using `Textnet.project`, a `connected` argument can also be passed.
* Adds `TextnetBase.save_graph` to save the underlying graph (for instance, for further processing in Gephi).
* Improved and extended documentation and docstrings.
* Update dependencies.

This release is an attempt to fix the cross-platform build and deploy pipeline (to ensure binary wheels for all platforms are uploaded to PyPI). Otherwise, there is no substantive change from previous release.