fix the mismatch between gpu-manager pick up and gpu-admission predicate

[HeroBcat]

|---PHB (aval: 2, pids: [], usedMemory: 0, totalMemory: 35162423296, allocatableCores: 0, allocatableMemory: 0)
|   |---GPU0 (pids: [], usedMemory: 0, totalMemory: 11721506816, allocatableCores: 0, allocatableMemory: 0)
|   |---GPU1 (pids: [], usedMemory: 0, totalMemory: 11721506816, allocatableCores: 100, allocatableMemory: 11721506816)
|   |---GPU2 (pids: [], usedMemory: 0, totalMemory: 11719409664, allocatableCores: 100, allocatableMemory: 11719409664)

In my case, the total memory of GPU2 (11719409664) is less than GPU1 (11721506816).
So that even if the two GPUs are not allocated, gpu-manager will pick up GPU2 and gpu-admission will predicate GPU1, which leads to mismatch.
We can divide AllocatableMemory by 256 MB (types.MemoryBlockSize) first. And then compare the results.

[mYmNeo]

Thank you for your commit, could you append a test for this case?

[HeroBcat]

Thank you for your commit, could you append a test for this case?

I have updated the test case, please check.