feat(event-dispatcher): add async in-process event bus

[pedro-fs-garcia]

Implement EventDispatcher core module to decouple domain emitters from consumers, avoiding circular dependencies and growing coupling between domains (chatbot, ticket, live_chat).

Summary

• Introduce an async in-process EventDispatcher in app/core/event_dispatcher/ to decouple cross-domain side effects (e.g. triage completion triggering ticket creation and conversation opening) without external infrastructure (Kafka, Redis, etc.)
• Define typed Pydantic payload schemas and an AppEvent enum catalog for 6 domain events: triage.finished, ticket.created, ticket.assignee_updated, ticket.status_updated, ticket.escalated, ticket.closed
• Enforce handler correctness at startup via @event_handler decorator validation and EVENT_PAYLOAD_MAP wiring — schema mismatches raise InvalidHandlerError before the app serves traffic
• Handlers run as independent asyncio.Task (fire-and-forget) with structured exception logging, so a failing listener never blocks the emitter or other listeners
• Wire dispatcher singleton into FastAPI lifespan (main.py) with a register_app_events_listeners hook ready for domain listeners

Motivation

As business flows grow (triage → ticket → conversation → notification), injecting services across domains creates circular dependencies and violates domain boundaries. The dispatcher inverts this: emitters publish events without knowing who reacts, and new side effects are added by subscribing listeners — no changes to the originating service.

See docs/event_dispatcher.md for the architectural decision, event catalog, chained event flow, and usage guidelines.

Test plan

• Unit tests for EventDispatcher.subscribe and publish — valid handlers, duplicate subscription idempotency, payload type mismatch rejection
• Unit tests for @event_handler decorator — payload validation, error logging on handler failure, schema error propagation
• Verify app starts without errors and logs Registering event listeners to EventDispatcher.
• Confirm no regressions on existing endpoints

🤖 Generated with Claude Code