Improve the performance of getDistinctPartitionReadFileFormats for HiveTableScanExecTransformer

### Description

Currently, the original code has a lot of issues. 

- Perform expensive `HiveClientImpl.fromHivePartition()` operation on each partition

- Perform two traversal using exists() and map()

- No caching mechanism, repeated calculation of the same format conversion

### Gluten version

main branch