Back up tool_test_output.* rather than overwriting

The old file could be renamed to `tool_test_output_backup_${n}.json` or similar. Or more simply, a timestamp could be added to every file name.

Locate the code path that writes tool_test_output.* and trace how repeated runs handle an existing file. Decide between numbered backups and timestamps based on the surrounding behavior; done means reruns preserve earlier output instead of overwriting it.

Written by the indexing model from the issue text.