[datlechin]

Summary

• a_million_row_table_pages_to_its_end_and_exports: where the server pages, the last page may now take twice the export's time, not once.
• Skipping to the last page and the export both read every row at the server's speed. On ScyllaDB the skip took 0.89 to 1.04 times the export over the last ten CI runs, so the case failed on noise: the last page took 5.87s, past 5.67s on fix(linux): list Redis and DuckDB in the desktop entry, metainfo and man page #3823, 8.3s past 8.0s on feat(linux): add the AI assistant #3822.
• Cassandra sits near 0.5 times, so the bound still catches a skip that reads the table more than twice.
• docs/testing.md says so.

Test plan

• CI's Docker tests (tablepro-driver-cassandra) run the case on Cassandra and ScyllaDB.
• cargo clippy -p tablepro-test-support --all-targets -D warnings.