Add AI NPC believability smoke tests and evidence pack

[JOY]

Summary

Turn the AI NPC believability evaluation contract into runnable Play Mode and backend smoke tests.

Why

The design now defines a concrete prompt pack and evidence criteria for checking whether NPCs differ by role, memory, relationship, state of mind, hidden-lore boundaries, authority safety, and first-session staging. The next step is to make this repeatable instead of relying on ad hoc screenshots.

Acceptance Criteria

• Add a Play Mode smoke path for the believability prompt pack across at least three focus NPCs.
• Use the concrete prompt ids and expected role directions from docs/design/108-alpha-ai-npc-believability-prompt-pack.md.
• Add a Nakama test fixture with two NPC profiles, one memory event, and one relationship delta.
• Verify one before-memory and one after-memory response path.
• Verify hidden body-transfer lore stays scoped to NPC knowledge.
• Verify first-session NPC choreography: barks stop during focused dialogue, outcome reaction waits for outcome state, and no NPC grants reward/item/proof/body access.
• Export or display redacted PromptTrace summaries for the tested turns.
• Add an anti-repeat regression check for focused chat and NPC-to-NPC speech.
• Save evidence using the public-safe evidence pack rules from docs/design/53-ai-npc-believability-evaluation.md, docs/design/93-alpha-ai-npc-believability-smoke-evidence-plan.md, and docs/design/108-alpha-ai-npc-believability-prompt-pack.md.

References

• docs/design/53-ai-npc-believability-evaluation.md
• docs/design/93-alpha-ai-npc-believability-smoke-evidence-plan.md
• docs/design/101-alpha-focus-npc-first-session-choreography.md
• docs/design/108-alpha-ai-npc-believability-prompt-pack.md
• docs/design/34-alpha-acceptance-matrix.md
• docs/design/37-ai-npc-backend-client-roadmap.md