feat: Support spaCy 3.7-3.8; Python 3.12

[DanielSchaffer]

Description

• Added support for Python v3.12.
• Added support for spaCy v3.7 and v3.8.
• Added models for English, French, German, and Polish for spaCy v3.7/v3.8
• Updated French language rules for spaCy v3.7/v3.8

Internal changes and fixes

• Added documentation and agent skills for adding support for new spaCy versions.
• Added conditional use of importlib.resources instead of pkg_resources for Python >= 3.9.
• Migrated package setup to pyproject.toml.
• Fixed outdated URLs for training corpora.
• Improve scripts for downloading corpora and converting French corpora to conll.
• Bumped version to 1.5.0

Addresses #29

---

I'm not sure if anyone is still monitoring this repo. This isn't quite ready to merge, but I'd appreciate if someone could start taking a look, as there were some changes necessary specifically to support spaCy v3.7+

TODO

• Finish fixing failing tests for French models
• Validate v3.8 functionality
• Figure out management of the model .zip files
• French speaker validation of rule adjustments

Model .zip files

The en zip now exceeds 100MB, which is GitHub's limit for public repositories. Beyond that, you'd need to use git-lfs, but I'm not sure what's involved in setting that. I don't think it's free. Are they used for anything beyond a convenience download?

[mjspeck]

Is there any hope of this actually being merged? The last commits on master were 3 years ago...

[DanielSchaffer]

@mjspeck

Is there any hope of this actually being merged? The last commits on master were 3 years ago...

Not sure. I'm going to see if I can contact the author directly. Failing that, I can try publishing a separate package. You're welcome to give it a sanity check :)

[richardpaulhudson]

Hi @DanielSchaffer, thankyou very much for all your work! I'm no longer actively working on Coreferee, but I'm glad people are finding it useful and will definitely do my best to get this merged.

On the individual points you raise:

• for the zip file size issue, I would suggest removing the earlier spacy versions from the latest zip files to bring the size back down to below the GitHub threshold. I doubt anyone would have any reason to continue using the old spaCy models, and even if anybody does they can still download an earlier version of the zip file in question.
• With respect to the failing French tests, I'm afraid I'm probably not the right person to help with that. While I wrote the English, German and Polish rules myself, the French rules and models were submitted by @Pantalaymon once Coreferee had already been released. Unfortunately my knowledge of French is not really extensive enough to enable me to maintain them with any confidence, which is why even before this PR the French support was not as up-to-date as the support for the other three languages. Perhaps someone else can help, otherwise we would need to decide between merging with issues or removing the French support from the PR.
• You write 'validate 3.8 functionality', what exactly are the open questions here?

[DanielSchaffer]

@richardpaulhudson

Zip files make sense. For 3.8, I've actually resolved that - I just hadn't had the chance to make sure tests were passing on both 3.7 and 3.8, but that's done now.

For French, my colleague @eugenielm has offered to take a look, though if @Pantalaymon has some time to review that would be great as well.