Parallelizing knn and inrange searches

[SebastianAment]

Overview

This PR uses the @threads macro to parallelize inrange and knn searches over independent datapoints.

Benchmarks

Setup

using NearestNeighbors
using BenchmarkTools

d, n = 100, 10_000;
X = randn(d, n);
T = BallTree(X);

On Master

@btime knn(T, X, 3);
  6.150 s (20008 allocations: 9.31 MiB)
@btime inrange(T, X, 5);
  6.271 s (20005 allocations: 9.08 MiB)

With this PR

@btime knn(T, X, 3);
  766.583 ms (20090 allocations: 9.32 MiB)
@btime inrange(T, X, 5);
  797.295 ms (20087 allocations: 9.09 MiB)

This is a factor of 8 improvement, on a 2021 MacBookPro with an M1 Pro and 16 threads.