Refactor: MovingInterquartileMean

[Swiftb0y]

what was initially just supposed to be a cleanup/refactor, mutated into a full blown optimization. This is primarily intended as a cleanup though, and I only implemented the benchmark to show that this was not causing regressions in performance. There is still a little more to be done, but these are some preliminary numbers.
Before:

-----------------------------------------------------------------------------
Benchmark                   Time             CPU   Iterations UserCounters...
-----------------------------------------------------------------------------
BM_insertion/16         17485 ns        16857 ns            1 items_per_second=3.79664M/s
BM_insertion/64         20922 ns        20926 ns            1 items_per_second=12.2336M/s
BM_insertion/512       774264 ns       770344 ns            1 items_per_second=2.65855M/s
BM_insertion/4096    55161004 ns     54684805 ns            1 items_per_second=299.608k/s
BM_insertion/16384  891138337 ns    886767771 ns            1 items_per_second=73.9044k/s

After:

-----------------------------------------------------------------------------
Benchmark                   Time             CPU   Iterations UserCounters...
-----------------------------------------------------------------------------
BM_insertion/16         10458 ns         9738 ns            1 items_per_second=6.57219M/s
BM_insertion/64         16030 ns        15969 ns            1 items_per_second=16.0311M/s
BM_insertion/512       745585 ns       735096 ns            1 items_per_second=2.78603M/s
BM_insertion/4096    40907286 ns     40517411 ns            1 items_per_second=404.369k/s
BM_insertion/16384  644210901 ns    641245320 ns            1 items_per_second=102.201k/s

[Swiftb0y]

So I added yet another optimization (custom replace_sorted algorithm instead of erase+insert) which adds further speedups, especially with realistic IQM sizes.

Edit: Just realized that the Y-Label is not quite right, its "elements processed per second".
Edit2: The error bars denote $\pm1\sigma$ (standard deviation) to show that the results are reasonably significant

Please review.

[Swiftb0y]

I decided to drop replace_sorted as it introduced quite a lot of extra code without much benefit. This should now quite drastically simplify the code while still providing a speedup.

[JoergAtGithub]

Wouldn't
std::queue<double>().swap(m_queue);
work too? Not tested.

[Swiftb0y]

good catch, yeah that should work.

[JoergAtGithub]

LGTM! Thank you!

[Swiftb0y]

just noticed I still had some unpushed commits in my tree, it would be nice if could take a look at those too. Thank you.

[JoergAtGithub]

LGTM! Please squash the fixups!

[Swiftb0y]

Done. Thank you.

[JoergAtGithub]

Thank you!

[Swiftb0y]

Thank you for the review.