Multiple conversations

Attach MultipleConversationsExample and AstralRuntimeInitializer to a GameObject. Configure a generation model before starting the two independent NPC conversations through one model executor.

Each conversation owns its native slot and stream buffer. The coroutine polls every active slot once per frame, so one quiet NPC does not block the others. Cancel uses generic request references, and teardown releases conversations before the shared model.

Set modelPath and backendName for a real generation model. Keep maxBatchTokens within the model's configured batch capacity.