[BUG] [Graphscope on yarn] When generating Fragments, there is a partition issue.

[rouxiaomin]

Describe the bug
1.I have 3 machines.The initialization of executors in Spark is evenly distributed across different machines.

2.When starting to generate the metadata for the FragmentRDD，the metadata information is skewed to one particular machine

3.After analysis, the problematic code is likely to be here。There is an issue with the initialization of variables rddhosts、rddLocations and partitionIds. The partition acquisition mechanism here needs to be checked.