Closes #5250

read_parquet() SQL function now supports parallel execution. This means that parallel filtering and aggregation supported by QuestDB's query engine now apply to read_parquet() as if it was a native table.

Note: we don't support projection for virtual table functions like read_parquet(), so if you have, say, 1000 columns in your parquet, but you query only a single column, we'll end up decoding all 1000 columns. This is something to improve in future. Tracked in #5280

Parallel read_parquet() execution can be disabled with new cairo.sql.parallel.read.parquet.enabled configuration property.

Also includes the following:

• Removes redundant methods from the PageFrameCursor interface
• Fixes incorrect error handling in parquet-to-native partition conversion

TODOs:

• Try extracting physical table-only methods from PageFrameCursor into a child TablePageFrameCursor interface and use it where appropriate.
• Add a prop to enable/disable parallel read_parquet().