Implement a dictionary-based nickname generation system for Nuzlocke runs. Instead of using an LLM API to generate names on the fly, provide a static dictionary of words categorised by theme. A word can belong to multiple categories, making it usable across different naming schemes.
- **Dictionary storage**: Static JSON file in `backend/src/app/seeds/data/`, alongside other seed data. Not exposed to frontend directly.
- **Dictionary format**: Category-keyed structure (`{ "mythology": ["Apollo", ...], "space": ["Nova", ...] }`) for fast lookup by naming scheme. Words may appear in multiple categories.
- **Suggestion logic**: Backend service with API endpoint. Frontend calls the backend to get suggestions.
- **Used-name tracking**: No new storage needed. The existing `Encounter.nickname` field already tracks assigned names. The suggestion engine queries encounter nicknames for the current run and excludes them.
- **Naming scheme per run**: Dedicated nullable `naming_scheme` column on `NuzlockeRun` (not in the `rules` JSONB).
- **~150-200 words per category**: A typical Nuzlocke has ~100 encounters, so this provides ample variety without repetition.
- **Name suggestion UX**: When registering a new encounter, the user is shown 5-10 suggested names from their chosen naming scheme. They can click one to select it, or regenerate for a fresh batch.
- **Naming scheme selection**: Users pick a naming scheme (category) per run, either at run creation or in run settings.
## Success Criteria
- [ ] Word dictionary data file exists with multiple categories, each containing 150-200 words
- [ ] Name suggestion engine picks random names from the selected category, avoiding duplicates already used in the run
