feat: Implement pronunciation store with SQLite backend

- Added a new module for managing pronunciation overrides using SQLite.
- Implemented functions to load, save, search, and delete pronunciation overrides.
- Introduced schema for storing overrides and metadata.
- Added thread-safe access to the database with RLock.
- Created a utility for normalizing tokens for consistent storage and retrieval.

refactor: Overhaul entities step in the preparation wizard

- Renamed Step 3 from "Speakers" to "Entities" across all templates and routes.
- Introduced sub-navigation with tabs for "People", "Entities", and "Manual Overrides".
- Enhanced UI to display detected entities and allow manual overrides for pronunciations.
- Implemented search functionality for manual overrides with AJAX support.
- Updated frontend logic to manage tab interactions and voice selections.

docs: Add detailed plan for entities step overhaul

- Documented requirements, implementation strategies, and testing plans for the entities step.
- Outlined the integration of POS tagging and entity recognition using spaCy.
- Provided a comprehensive overview of the manual overrides workflow and data persistence strategies.

pyproject.toml

@@ -17,6 +17,7 @@ dependencies = [
     "misaki[zh]>=0.9.4",
     "ebooklib>=0.19",
     "beautifulsoup4>=4.13.4",
+    "spacy>=3.5,<4.0",
     "en-core-web-sm @ https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl",
     "PyMuPDF>=1.25.5",
     "platformdirs>=4.3.7",