fix(parquet/pqarrow): decoding Parquet with Arrow dict in schema

[freakyzoidberg]

Fix spurious parquet: column chunk cannot have more than one dictionary. with specific parquet file
Resolve #546

Parquet with

• Arrow Dict column
• Arrow Schema serialied in Parquet Metadata
• ColumnChunks with 1 dict page + at least 2 Data page

Bug

When maybeWriteNewDictionary() resets newDictionary = false at line 965, it causes the next call to readDictionary() to try to read the dictionary page again from the pager, which then calls configureDict() again, which throws the "cannot have more than one dictionary" error!

The sequence is:

1. Read DICTIONARY_PAGE → newDictionary = true
2. Read DATA_PAGE_1 → calls maybeWriteNewDictionary() → resets newDictionary = false
3. Read DATA_PAGE_2 → calls readDictionary() → since newDictionary = false, tries to get dictionary page again → calls configureDict() → ERROR because decoder already exists

Fix

Added DictionaryState enum (column_reader.go):

• DictNotRead: Dictionary page hasn't been read yet
• DictReadNotInserted: Dictionary page read and decoder configured, but not inserted into Arrow builder
• DictFullyProcessed: Dictionary fully processed (read + inserted into builder)

[ahmed-mez]

Thanks for the fix and reproducer @freakyzoidberg !

I confirm that running your branch on the file I mentioned in #546 now doesn't return the parquet: column chunk cannot have more than one dictionary. error any more.

[zeroshade]

Only one critique, everything else looks great! 😄 Thanks!

[zeroshade]

my only critique is that we shouldn't export these.