Speedup for bounding_box_query

[MeyerBender]

Speedup for bounding_box_query(), especially in the case of many points in the object. Main improvements:

• removed unnecessary compute() call
• rewrote transformation logic: in the case of identity or scaling transformations, we can avoid the two-step procedure of doing a coarse query in intrinsic space and then a fine-grained on in the global coordinate space

With this, we achieved speedups between 30% and 70% (depending on the data). In the future, one could consider using a spatial index for the point to speed things up further.

@LucaMarconato it would be great if you could add a test for the case we discussed at the hackathon, just to ensure our rewrite of the transformation logic didn't break anything.

CC @dschaub95

[codecov]

Codecov Report

❌ Patch coverage is 96.61017% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.06%. Comparing base (ea87931) to head (561f5f3).
⚠️ Report is 18 commits behind head on main.

Files with missing lines	Patch %	Lines
src/spatialdata/_core/query/spatial_query.py	96.55%	2 Missing ⚠️

Additional details and impacted files

@@ Coverage Diff @@## main #1104 +/- ##
==========================================
+ Coverage 92.04% 92.06% +0.01% 
==========================================
Files 51 51 Lines 7785 7792 +7 ==========================================
+ Hits 7166 7174 +8 + Misses 619 618 -1 

Files with missing lines	Coverage Δ
src/spatialdata/models/_utils.py	89.20% <100.00%> (+0.06%)	⬆️
src/spatialdata/_core/query/spatial_query.py	95.52% <96.55%> (+0.33%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[LucaMarconato]

_bounding_box_mask_points() is called only in two locations, and in both points_df is available. I will therefore remove the points argument, and only keep points_df.

[LucaMarconato]

This function uses single dispatch and expects a Dask DataFrame. I will extend this to support also pd.DataFrame. In the future we will allow points to be pd.DataFrame also inside the SpatialData object.

[LucaMarconato]

Thanks! This is the test case I wanted to add (it's a case in which the bounding box of the backward-transformed bounding box contains extra points with respect to the backward-transformed queried points).

Here the queried point is C, and the bounding box of the backward-transformed bounding box would contain A, B, C.

The test passes!