Cache gpu_alloc_map in Redis, and Add RescanGPUAllocMaps mutation

[jopemachine]

Motivation  

• Since gpu_alloc_map exists on the agent, querying this field requires an RPC call.
  In a production environment with multiple agents, repeatedly querying this field is inefficient and can significantly slow down response time.

Required Features

• Let's cache the gpu_alloc_map in Redis and create a new mutation based on a background task to update the cache.

Impact  

• This will allow us to monitor the GPU usage of all agents with less overhead (while updates are necessary for an accurate understanding of the situation)

Testing Scenarios  

• We need to ensure that the background task is correctly triggered after the cache update mutation is executed and verify that the gpu_alloc_map for all agents is cached once the task is completed.