Optimize dandiset listing endpoint

[jjnesbitt]

This reduces the load time of the dandiset listing endpoint by optimizing the _get_dandiset_to_version_map function. This means the search endpoint should also experience a slight performance improvement, since it also uses that function.

The optimization is accomplished with the following changes:

1. Removes an unnecessary query being made in the get_root_paths_many function
2. Shifts the version stats aggregation to be done in postgres, making it more efficient
3. Optimize the query for retrieving the latest version. Similar to the other optimized query, this leverages postgres aggregation to efficiently return the latest published versions for all relevant dandisets
4. Use two smaller queries instead of one union-ed query when retrieving drafts and published versions, to take advantage of the last published version optimization.

[danlamanna]

This LGTM, it seems like a performance improvement and a code simplification which is nice.

[dandibot]

🚀 PR was released in v0.3.59 🚀