Implement box filtering

[camposandro]

Implements box filtering for range of right ascension (ra) and/or declination (dec). If both are provided we construct a spherical polygon and use the preexisting polygon search. This feature allows to select the partitions of a subregion of the catalog sky, reducing the number of partition files read and therefore improving runtime performance.

Needed for astronomy-commons/lsdb#90.

• My PR includes a link to the issue that I am addressing

Code Quality

• I have read the Contribution Guide
• My code follows the code style of this project
• My code builds (or compiles) cleanly without any errors or warnings
• My code contains relevant comments and necessary documentation

[github-actions]

Before [fb575dd]	After [fd7498a]	Ratio	Benchmark (Parameter)
650±10ms	657±9ms	1.01	benchmarks.Suite.time_pixel_tree_creation
96.7±1ms	97.7±0.5ms	1.01	benchmarks.time_test_cone_filter_multiple_order
1.15±0.01s	1.15±0.03s	1	benchmarks.MetadataSuite.time_load_partition_info_order7
1.87±0.01s	1.86±0.02s	1	benchmarks.MetadataSuite.time_load_partition_join_info
126±3ms	125±2ms	0.99	benchmarks.time_test_alignment_even_sky
305±7ms	294±10ms	0.96	benchmarks.MetadataSuite.time_load_partition_info_order6

Click here to view all benchmarks.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (fb575dd) 100.00% compared to head (10945ca) 100.00%.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #202   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           54        55    +1     
  Lines         1848      1923   +75     
=========================================
+ Hits          1848      1923   +75     

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

[hombit]

It feels and works great.

One note about performance: selecting large range, in any of ra or dec, for norder=10 takes around a minute on my machine. I believe it is the same order of magnitude as other actions on trillion-row scale catalogs, so maybe it is not a big issue for now.

[smcguire-cmu]

It feels and works great.

One note about performance: selecting large range, in any of ra or dec, for norder=10 takes around a minute on my machine. I believe it is the same order of magnitude as other actions on trillion-row scale catalogs, so maybe it is not a big issue for now.

I'd be interested in seeing where that time is. My guess would be creating the pixel trees for large areas at order 10. Not needed now, but I wonder if we convert our large order 10 area to a MOC if that would make it quicker by reducing the number of pixels in the tree. We could also improve the pixel tree performance ofc (or rewrite it in rust) but that might be more difficult.