We are currently running this migration in production to reindex all the
notes in GitLab.com. There happen to be over 30M of them so it will take
quite a while to run. We have found in
https://gitlab.com/gitlab-org/gitlab/-/issues/324745 that the migration
is not performing as well as we'd hoped. Given that we want to keep each
job taking less than 1 minute to run we want to reduce the batch size to
make that easier.

We have found that 9000 per 3 minutes is too much and we are regularly
overlapping with the previous batch and thus we end up often processing
the same batch twice.

In addition it will be more efficient to push the work into the "initial
indexing queue" instead of our main "incremental indexing queue" as the
initial indexing queue has far less traffic (it only covers newly
created projects while incremental is any record update in any project).