Fix flaky checkpoint continuation: sys__id not preserved when copying partial output table

[ilongin]

When a mapper UDF crashes mid-execution and the user re-runs with a fix, _continue_udf copies the parent partial output table into a new table. During this copy, insert_into strips sys__id and auto-generates new ones. If the rows end up with different sys__id values than the originals, things break because:

• The mapper explicitly preserves input sys__id in its output (1:1 mapping)
• create_result_query joins the output table with the input table on sys__id
• calculate_unprocessed_rows uses sys__id to determine which inputs were already processed

When the copied IDs don't match, the join produces wrong result-to-input pairings — values get swapped, duplicated, or lost.

The bug has existed since UDF checkpoints were introduced.