Use std::threads instead of openmp for parallelizing qsort

[r-devulap]

First stab at switching to std::threads to not have the dependency on openmp. These look like about 10-20% slower than the openmp version, not entirely sure why:

Benchmark                                                                     Time             CPU      Time Old      Time New       CPU Old       CPU New
----------------------------------------------------------------------------------------------------------------------------------------------------------
[simdsort/random_10m/int* vs. simdsort/random_10m/int*]64_t                +0.1881         -0.8570      36212571      43022493      36211824       5177321
[simdsort/random_10m/int* vs. simdsort/random_10m/int*]32_t                +0.1628         -0.8453      16560818      19257370      16560817       2561878
[simdsort/random_10m/int* vs. simdsort/random_10m/int*]16_t                +0.1600         -0.8628      11944131      13854631      11942415       1638489
OVERALL_GEOMEAN                                                            +0.1702         -0.8552             0             0             0             0

[r-devulap]

I know, this needs a new macro.

[Copilot]

Pull Request Overview

This PR eliminates the dependency on OpenMP by replacing parallelized qsort code with a std::thread implementation using a dedicated threadmanager.

• Updated qsort_ and related functions to use std::thread and threadmanager instead of OpenMP tasks.
• Removed conditional OpenMP code blocks and adjusted thread management in both common and AVX512-specific qsort headers.

Reviewed Changes

Copilot reviewed 3 out of 5 changed files in this pull request and generated 1 comment.

File	Description
src/xss-common-qsort.h	Replaced OpenMP tasks with std::thread calls and threadmanager usage.
src/xss-common-includes.h	Added necessary thread and mutex includes; introduced threadmanager struct.
src/avx512-16bit-qsort.hpp	Updated qsort calls to use threadmanager instead of OpenMP threshold.

Files not reviewed (2)

• Makefile: Language not supported
• scripts/bench-compare.sh: Language not supported

[Copilot]

Setting max_thread_count to 0 in non-OpenMP builds disables parallel thread spawning. Consider initializing it with a non-zero value (e.g., based on std::thread::hardware_concurrency()) to enable proper multithreading.

[r-devulap]

Closing this in favor of #201