Introduce nearest neighbor search for points

[ullingerc]

QLever now supports a special predicate ?left <nearest-neighbors:k> ?right where k is a positive integer that is transformed into a spatial join operation which for each entry in ?left only returns the k spatially closest elements in ?right.
For the semantically correct usage, it is important that the <nearest-neighbors> join is performed as the last step, e.g. by putting the other constraints for ?left and right into subqueries. In the future we probably want to replace this special predicate by a special construct that better reflects the non-commutativity of the nearest neighbor join.

Currently, the nearest neighbor join (same as the <max-distance-in-meters> join only work for point objects). The join is implemented using Google's S2 library, delivers the functionality for nearest neighbor search out of the box.

[codecov]

Codecov Report

Attention: Patch coverage is 95.85799% with 7 lines in your changes missing coverage. Please review.

Project coverage is 88.40%. Comparing base (b97c44c) to head (c4be461).
Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/engine/SpatialJoin.cpp	96.22%	1 Missing and 5 partials ⚠️
src/engine/QueryPlanner.cpp	66.66%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1528      +/-   ##
==========================================
+ Coverage   88.33%   88.40%   +0.07%     
==========================================
  Files         362      362              
  Lines       27319    27426     +107     
  Branches     3682     3705      +23     
==========================================
+ Hits        24131    24245     +114     
+ Misses       1952     1945       -7     
  Partials     1236     1236              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[joka921]

This already looks very very promising and is also working like a charm (apart from a serious lifetime bug, which we fortunately haven't triggered so far:)).

I have left you an initial round of comments,
So you can work on improving this really really nice feature.

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarCloud

[joka921]

Thank you very much, I think we can merge this now.

[hannahbast]

@ullingerc This is now active on our OSM Planet instance. I just noticed a small problem, which might have an easy fix. For example, consider https://qlever.cs.uni-freiburg.de/osm-planet/N1UUCA . QLever should compute the result of the subquery first (which contains the magic <nearest-neighbors:1> predicate), but it does not. The reason seems to be the huge size estimate for the result of the nearest neighbor join (3,517,552,729,752 according to the "Analysis" tree). Shouldn't the size estimate for this join simply be the size estimate of the left side?