query() with release_status="any" fetches every match twice, truncates silently, and loses the sort order #176

Summary

KGClient.query() with release_status="any" (used by KGObject.list(..., release_status="any") and by every existence query in save() and exists()) is implemented by fetching everything and paging in Python:

1. it runs the query against the released stage with size=100000, ignoring from_index and size;
2. it runs it again against the in-progress stage, also with size=100000;
3. it merges the two by id, the in-progress version replacing the released one, and only then applies from_index and size.

This has three consequences:

• Cost. Asking for one page of results fetches every matching instance, twice. For a broad query on a large type this can mean very large responses, or a timeout, to return a handful of results.
• Silent truncation. If either stage has more than 100000 matches, the extra ones are dropped without any warning, and total is wrong.
• Ordering. Each stage's results are sorted by the class's sort_property, but the merge keeps the released results in their order and appends the in-progress-only instances at the end. So with "any", results are not sorted as a whole, and paging through them with from_index gives an order that depends on release status.

The code has a comment acknowledging that this is "simple but very inefficient".

Proposal

• Page through both stages together (a merge of two sorted streams by sort_property), fetching only as many results from each as are needed for the requested page (at most from_index + size from each). Instances present in both stages need deduplicating, and may sort differently in each if a name-like property has changed since release.
• At the least: raise or warn instead of silently truncating at 100000, and sort the merged list by sort_property before applying from_index and size.