Revisit dropped Partition Columns handling

Currently, dropping partition columns (controlled by "hoodie.datasource.write.drop.partition.columns") is handled in a piecemeal fashion, which unfortunately may to lead to very subtle and hard to troubleshoot issues when used.

For ex, currently in HoodieSparkSqlWriter this would affect what will be persisted as writer's schema – in case partition columns are dropped from the data file we will persist "reduced" schema as the one that was used by the Writer, which is invalid since Writer was using the full schema, however partition columns weren't persisted in the Data Files (ie dropped, since they're already encoded into the partition path)

## JIRA info

- Link: https://issues.apache.org/jira/browse/HUDI-4772
- Type: Bug
- Affects version(s):
- 0.13.0

Start at HoodieSparkSqlWriter and trace how the hoodie.datasource.write.drop.partition.columns setting affects the writer schema and persisted data-file schema. Done means the writer uses the full schema while partition columns remain omitted from data files and are represented in the partition path.

Written by the indexing model from the issue text.