Distinguish individual characters from groups/factions

[renaudcepre]

Problem

The analyzer currently extracts collectives as characters: "the orcs", "the dwarves", "the drones", "the Nazgûl". These pollute the character graph with non-character nodes that have no meaningful profile.

Observed in:

• LotR corpus: The Orcs, The Dwarfs, Black Riders extracted as characters
• Felix corpus: the drones, La Canopée confused with Andrew Milton

Short-term fix

Update CHARACTER_PROMPT in analyzer.py to explicitly exclude races, armies, factions, and unnamed collectives. Only individually named characters should be extracted.

Long-term vision

Introduce a Group node type distinct from Character:

```
(Gimli) -[:MEMBER_OF]-> (:Group {name: "Dwarves"})
(Aragorn) -[:MEMBER_OF]-> (:Group {name: "Fellowship of the Ring"})
```

This would enable:

• Filtering characters by faction/race
• Querying "which characters are Elves"
• Group-level arc tracking

Open questions

• Where is the line? "The Nazgûl" as a force vs a named Nazgûl individual
• Should the LLM infer MEMBER_OF relations, or should it be a separate extraction step?
• Should groups appear in the beat graph (subject/object)?