Refactor create_kernels RPC call to asynchronous event-driven pattern

[fregataa]

Current Behavior

The "manager" component makes a RPC call create_kernels to "agent" components. The manager blocks and waits for the RPC response from agents, which:
Creates tight coupling between managers and agents
Increases latency as managers must wait for kernel creation completion

Proposed Changes

Refactor the create_kernels RPC function to use a fire-and-forget pattern with event-based result handling:

1. Manager component:

• Implement event subscription to receive kernel creation failure (managers are already handling creation success events).

2. Agent component:

• Modify create_kernels RPC handler to spawn background coroutine tasks
• Return response in create_kernels RPC function immediately
• Add error handling and dispatch kernel creation failure events