feat(ai-service): Replace task_results global dict with Redis-backed task store

[AugistineCreates]

Description

This PR replaces the module-level in-memory dictionary task_results in the AI Service (app/ai-service/tasks.py) with a persistent, Redis-backed task store.

Why this is necessary:

• Resilience: A single-process in-memory dict gets destroyed whenever a pod/container restarts, losing active status tracking.
• Scalability: In multi-process deployments, workers run in forked/different processes and cannot share status updates via a local in-memory dict, causing fork-bomb state mismatches.

Changes Made

• Removed task_results: Deleted the task_results global dictionary.
• Lazy Redis Client: Added lazy initialization of the Redis client (get_redis_client) targeting the configured REDIS_URL.
• Status Persistence: Implemented tasks.set_status to write task payload states directly to Redis with a TTL of 24 hours (86400 seconds) matching Celery's result_expires.
• Robust Retrieval: Updated tasks.get_task_status to check Celery's AsyncResult status first, and fallback to the Redis-backed store on failure or missing keys.
• Unit & Integration Tests: Created app/ai-service/tests/test_tasks_redis.py covering:
  • Setting status to Redis with precise TTL.
  • Graceful fallback reading from Redis when Celery returns unfinished states or throws connection errors.
  • Correct E2E cross-process communication simulation (Process A writing to Redis, Process B observing status).

Acceptance Criteria Verified

• Redis storage and 24-hour TTL enforced.
• Reads query Celery AsyncResult first and fallback to Redis.
• E2E behavior verified (Process A updates state, Process B reads state).