Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

In parquet writer, emit created_by string in correct format, e.g. ClickHouse version 25.10.1 (build 5b1dfb14925db8901a4e9202cd5d63c11ecfbb9f) instead of ClickHouse v25.9.1.1-testing.
Fix parquet reader compatibility with bad files written by old parquet-mr.

This PR is stacked on top of #87600 , review only commits starting from "A few more parquet fixes".

Closes #87306

• Workaround for files written by old duckdb with incorrect data_page_offset.
• Make parquet writer emit created_by string in correct format (according to parquet.thrift): ClickHouse version 25.10.1 (build 5b1dfb14925db8901a4e9202cd5d63c11ecfbb9f) instead of ClickHouse v25.9.1.1-testing.
• In the old parquet reader, unbreak a workaround for another workaround to allow reading bad files written by old parquet-mr.
• Fixed clickhouse-test failing if diff output is not valid utf8.
• Replaced 00900_long_parquet_load.sh with 00900_long_parquet_load_2.sh that:
  • Doesn't use an additional python script and arrow's parquet-reader tool (which is broken: Fix parquet-reader outputting invalid json arrow#72 ).
  • Picks up newly added files automatically without needing to do some manual steps which no one knows about.
  • Doesn't double the number of files in data_parquet/.
  • Is simpler.
  • But: the .reference file needs to be updated whenever a file is added to data_parquet/ or schema inference logic changes.