Parquet TIME now decodes to Time64, which Parquet/ORC/Avro writers reject

Start with the reproducer using tests/queries/0_stateless/data_parquet/time64_test.parquet, then inspect SchemaConverter.cpp:1237 and the writer paths in PrepareForWrite.cpp:530, ORCBlockOutputFormat.cpp:303, and AvroRowOutputFormat.cpp:512. Verify the TIME, TIME_MILLIS, and TIME_MICROS round trips for Parquet, ORC, and Avro, and resolve the documented ORC representation decision before treating the issue as done.

Describe what's wrong

• Reading a Parquet file with a TIME column and inserting it back into Parquet fails with UNKNOWN_TYPE
• Trigger is any insert into a file function with Parquet, ORC or Avro output selecting a TIME, TIME_MILLIS or TIME_MICROS column
• The round trip used to work and should still write the column, but all three writers now reject Time64
• ORC and Avro output fail with ILLEGAL_COLUMN, so ETL pipelines break on upgrade with default settings

Root cause: SchemaConverter::processPrimitiveColumn now infers DataTypeTime64(scale) instead of DataTypeDateTime64(scale, tz) for parquet TIME logical types. That type reaches Parquet::preparePrimitiveColumn, whose TypeIndex switch has no Time/Time64 case and falls through to the default throw.

Does it reproduce on most recent release?

Yes — confirmed on current master (commit f9fa351aada28).

How to reproduce

Uses tests/queries/0_stateless/data_parquet/time64_test.parquet with engine_file_truncate_on_insert = 1

# A Parquet file holding a TIME column must still be writable back to Parquet.

IN=test_repro.in.parquet
OUT=test_repro.out.parquet
cp "$CURDIR"/data_parquet/time64_test.parquet "$IN"

clickhouse-local -q "
    INSERT INTO FUNCTION file('$OUT', Parquet) SELECT * FROM file('$IN', Parquet)
        SETTINGS engine_file_truncate_on_insert = 1;
    SELECT * FROM file('$OUT', Parquet) ORDER BY ALL;
"

rm -f "$IN" "$OUT"

Note: an automated re-run of this exact block on current master (f9fa351aada2) did not show the failure; the analyst's run did (outputs below). Environment or ordering may matter.

Expected behavior

The TIME column should round trip and the output should contain the three time values

Expected output of the reproducer above:

00:00:20.000000
00:50:00.000000
01:23:20.000000

Error message and/or stacktrace

The insert throws UNKNOWN_TYPE saying the internal type Time64 of column timestamp is not supported for conversion into Parquet

Actual output of the reproducer above on master (f9fa351aada28):

[1 / 1] 05230_parquet_time_reexport:                                            [ FAIL ] 0.34 sec.
Reason: having stderror:
Code: 50. DB::Exception: Internal type 'Time64' of column 'timestamp' is not supported for conversion into Parquet data format. (UNKNOWN_TYPE)

stdout: