Partition update with global index in MOR tables resulting in duplicate values during read optimized queries

The way Hudi handles updating partition path is by locating the existing record and performing a delete on the previous partition and performing insert on new partition. In the case of Merge-on-Read tables the delete operation, and any update operation, is added as a log file. However since an insert occurs in the new partition the record is added in a parquet file. Querying using `QUERY_TYPE_READ_OPTIMIZED_OPT_VAL` fetches only parquet files and now we have the case where 2 records for given primary key are present

## JIRA info

- Link: https://issues.apache.org/jira/browse/HUDI-1508
- Type: Bug

Start with the Hudi-1508 JIRA report and trace how partition updates in Merge-on-Read tables handle the delete in the old partition and insert in the new partition. Reproduce the case with a global index and a read-optimized query, then verify that a primary key is returned only once after the fix.

Written by the indexing model from the issue text.