Investigate possible export performance and memory usage improvements.

          I'd even be fine with merging this as is and promise to start on further refactoring right away with both ideas:

• Using a db iterator
• using qs.none()

Another thing (to think about later) that came to my mind: In the case of incremental, we now still include artifacts that are new to a repository version, but guaranteed to be in another (exported) repositories previous version. Could we safely skip them too?
Say the logic switching from:

$\displaystyle \bigcup_{\textrm{version}} \left( \textrm{version}.\textrm{artifacts} \setminus \textrm{prev}(\textrm{version}).\textrm{artifacts} \right)$

to

$\left( \displaystyle \bigcup_{\textrm{version}} \textrm{version}.\textrm{artifacts} \right) \setminus \left( \displaystyle \bigcup_{\textrm{version}} \textrm{prev}(\textrm{version}).\textrm{artifacts} \right)$

Originally posted by @mdellweg in https://github.com/pulp/pulpcore/pull/4161#pullrequestreview-1551702739

Start with the export implementation discussed in pull request #4161. Investigate the db iterator and qs.none() ideas, then assess whether incremental exports can safely exclude artifacts already present in earlier exported versions. Done means a clearly scoped performance change with export results preserved.