read_table_parquet raises KeyError when a column is absent from the stored astropy table_meta_yaml #20639

Description

Table.read(..., format="parquet") raises KeyError: '_healpix_29' on Gaia DR3's HATS-partitioned parquet files.

Root cause

In astropy/io/misc/parquet.py, read_table_parquet (lines 313–317 at v8.0.1, lines 332–336 on main):

header_cols = {x["name"]: x for x in meta_hdr["datatype"]}
for col in table.columns.values():
    for attr in ("description", "format", "unit", "meta"):
        if attr in header_cols[col.name]:

This looks up every column of the file in the table_meta_yaml header. Here the header was written by astropy, and a tool then added a column without rewriting the yaml: hats-import adds _healpix_29. Any column missing from the yaml makes the read fail outright, even though the data is fine. pyarrow reads the same file without complaint.

Suggested fix

Treat a column that has no header entry as having no stored attributes:

col_hdr = header_cols.get(col.name, {})
if attr in col_hdr:
    setattr(col, attr, col_hdr[attr])

Reproduction

1. Read any Gaia DR3 HATS partition file with astropy.table.Table.read(path, format="parquet") on astropy 8.0.1.
2. It raises KeyError('_healpix_29').
3. pyarrow.parquet.read_table(path) on the same file succeeds.

A minimal synthetic case that should hit the same lookup (not run; derived from the code above):

1. Write a Table to parquet with astropy.
2. Add a column with pyarrow, keeping the original schema metadata.
3. Read the file back with astropy.

Versions: astropy 8.0.1 (conda-forge), pyarrow from the same env, Linux aarch64 and x86_64.