Initial implementation for argpartial_sort

[matwey]

No description provided.

[matwey]

@hombit I need your help with this POC. The algorithm itself is very simple.

[codspeed-hq]

Merging this PR will not alter performance

✅ 47 untouched benchmarks
⏩ 4 skipped benchmarks¹

---

_{Comparing stable_sort (b0c8bd5) with master (39106f9)}

Footnotes

1. 4 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[hombit]

Looks good. It would be interesting to see performance vs np.argpartition, even for a single data point, e.g. len=100_000, k=100.

[hombit]

I believe we need pos + 1 here because it is what we have before we pop. It also may be simplified with the specialized constructor.

Suggested change

	let mut heap = BinaryHeap::new();
	heap.reserve(pos);
	let mut heap = BinaryHeap::with_capacity(pos+1);

[matwey]

I've implemented the benchmarks. For now it works ugly for large k values. I don't know why

[hombit]

Our CI doesn't support these arguments, please remove