The celery worker’s output is piped through 2>&1 | tee /tmp/delete-1.txt | grep -v basegenericmodel_post_delete_callback meaning the output you see here does not include basegenericmodel_post_delete_callback messages. This was done because there are 200 MB worth of those messages!
This was recorded with asciinema rec -i3. Even though this recording is only 1.5 minutes long, there’s an unseen gap of about 8 minutes while processing.
Outside of this recording, I ran the same process a second time but with --multiprocess instead of --include-children on the mprof command.