fix: multi-tenant schema context for worker task execution

[cdbartholomew]

Summary

• Background tasks (async retain, consolidation, reflections) fail silently in multi-tenant deployments.
• Root cause: The worker executes tasks without setting the tenant schema context, causing two cascading failures:
  1. The cancellation check in execute_task queries public.async_operations instead of the tenant's schema, finds no row, and skips the task as "cancelled" — even though it wasn't.
  2. Even if that were bypassed, _authenticate_tenant would throw AuthenticationError because background tasks have no API key/token.
• Observed symptom: Every async operation logs Skipping cancelled operation: <id> immediately after being picked up.

Changes

• poller.py: Pass task.schema into task_dict["_schema"] so the executor receives it
• memory_engine.py (execute_task): Pop _schema from task dict and set _current_schema before the cancellation check
• memory_engine.py (_authenticate_tenant): Skip tenant extension auth for internal=True requests when schema is already set
• memory_engine.py (task handlers): Use RequestContext(internal=True) for all 4 background task handlers
• task_backend.py: Add schema_getter for dynamic schema resolution in submit_task and wait_for_result
• http.py: Pass tenant_extension to WorkerPoller in create_app

Test plan

• Deploy to dev with multi-tenant config
• Trigger async retain — verify task completes instead of logging "Skipping cancelled operation"
• Trigger consolidation — verify it runs in the correct tenant schema
• Trigger reflection creation — verify background generation completes
• Verify single-tenant (no extension) deployments are unaffected