Confirm KV cache functions correctly when using LoRAs

[mayabar]

• Check real vLLM's behavior when running with multiple loras.
• Scenario to use: send same prompt to two different loras, check if BockStored send twice.
• Implement the same behavior in the simulator.