Optimizations to nearest neighbor

[aschampion]

Thanks for the great library! A few optimizations for our use case of querying hundreds of thousands of nodes in large (10-100K node) trees:

• Inline annotation on point indexing: surprised this is not happening automatically, but I receive consistent ~5% speedup with these.
• Optimizing min_max_dist_2 to use O(P::DIMENSIONS) multiplications rather than O(P::DIMENSIONS^2) multiplications. This doesn't do much for 2-D, but yields 15-20% speedup for our 3-D cases. Note that due to order of floating point operations, the result of the function is not identical to the previous version. If this is important, I have a slightly less elegant but same O() version I can substitute that gives identical results to the old version.
• Removing priority queue allocations from single nearest neighbor queries. When making many nearest neighbor queries, this has a large effect. For us, combined with the other optimizations this has a 1.5-1.8x speedup. It's also important for the performance of a follow-up PR I have go doing tree-to-tree all nearest neighbor queries.

[codecov-io]

Codecov Report

Merging #35 into master will decrease coverage by 0.64%.
The diff coverage is 74.28%.

@@            Coverage Diff             @@
##           master      #35      +/-   ##
==========================================
- Coverage   98.22%   97.57%   -0.65%     
==========================================
  Files          18       17       -1     
  Lines        1237     1239       +2     
==========================================
- Hits         1215     1209       -6     
- Misses         22       30       +8     

Impacted Files	Coverage Δ
rstar/src/point.rs	100.00% <ø> (ø)
rstar/src/algorithm/nearest_neighbor.rs	91.11% <60.86%> (-5.35%)	⬇️
rstar/src/aabb.rs	100.00% <100.00%> (ø)
rstar/src/params.rs	84.61% <0.00%> (-1.10%)	⬇️
...ar/src/algorithm/bulk_load/bulk_load_sequential.rs	96.77% <0.00%> (-0.11%)	⬇️
rstar/src/rtree.rs	98.57% <0.00%> (-0.03%)	⬇️
rstar/src/algorithm/rstar.rs	96.75% <0.00%> (-0.02%)	⬇️
rstar/src/algorithm/removal.rs	98.78% <0.00%> (-0.02%)	⬇️
rstar/src/algorithm/selection_functions.rs	100.00% <0.00%> (ø)
... and 3 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 34dbac5...4e69af8. Read the comment docs.

[Stoeoef]

Thanks a lot for your work on this!

I like all the changes, the performance benefits speak for themselves. I wish there was a "SmallHeap" package that would not require to implement the heap "on the fly", but we'll do what we have to do!.

Thanks a lot for upstreaming this!
Version 0.8.0 just got released, containing these changes.