fix(parquet/pqarrow): selective column reading of complex map column

[zeroshade]

Rationale for this change

Upstream fix for the issue identified in apache/iceberg-go#737. When reading maps with nested values using column indices for selective column reading, if the child fields of the map weren't in the list of indices there was a problem:

• Maps are represented in Parquet as a list of key-value structs (list<struct<key, value>>
• The struct MUST have exactly 2 fields (key and value) to be converted into a proper Arrow typed Map column
• When applying the column filtering, if only the key OR value field (but not both) were in the list of columns, the resulting child struct would only have 1 field
• As a result, the Map.validateData() method would fail with a panic of arrow/array: map array child array should have two fields.

What changes are included in this PR?

In pqarrow/file_reader.go leaf filtering is disabled when reading a map's key-value struct. This will ensure both the key and value columns are always read together, maintaining the required 2-field structure for map array.

Are these changes tested?

Yes a test case is added for the change.

Are there any user-facing changes?

This only affects map type reading with column filter selection ensuring correctness. The only change is that a failure mode has been eliminated.

[alexandre-normand]

Thanks for fixing this quickly, @zeroshade! This is very much appreciated.