SYS-1732: SpaCy NLP - director name extraction proof of concept

[ztucker4]

Implements SYS-1732

Adds a new script, spacy_experiment.py, demonstrating the use of spaCy to extract names for pre-extracted 245 $c data. The "small" and "medium" versions of spaCy's model are used. The script assumes the presence of the file f245c_directors.txt.

Since spaCy does not work with Python 3.13 (see discussion here), the Dockerfile has been updated to use a base Python image with Python 3.12. While the latest version of spaCy is 3.8.4, requirements.txt specifies version 3.7.5 due to incompatibilities with Mac M1 architecture in later versions (discussion here).

To test, rebuild the container (docker compose build) and run the script (docker compose run ftva_data python spacy_experiment.py). You should see terminal output describing the total number of entities. Two output files should be produced, output_core_web_sm.csv and output_core_web_md.csv. These are CSV files with three columns: the original 245 data, the names extracted by spaCy, and the non-name "entities" extracted by spaCy.

[akohler]

@ztucker4 Thanks for putting this together. Neither the Small nor Medium model is perfect (nor the Large, which I also loaded and tested), but overall Medium is best with this data:

Differences between Small and Medium: 27

Model: Small
Better than Medium: 8
Left word out of name: 7
Left name out completely: 11
Tokenized name wrong: 1

Model: Medium
Better than Small: 17
Left word out of name: 5
Left name out completely: 3
Tokenized name wrong: 2

A few small changes before merging: switching from ast to json (since we know the source of the sample data), and a more Pythonic way to flatten the list of lists into a list of strings.
Thanks --Andy