💪 add FPCS downsampling algorithm

[jonasvdd]

This PR ports the Feature-Preserving Compensated Sampling algorithm to tsdownsample

https://ieeexplore.ieee.org/document/10669793

DONE:

• tests for fpcs
• (compare with vanilla python / rust implementation)
• Nan handling
• add test for same behavior for nan- and non-nan aggregators when no nans are present in the data

TODO:

• .rs tests
• analyze current fpcs.rs file (and mainly the macro, as I separated the fpcs_inner_comp) function.

[codspeed-hq]

CodSpeed Performance Report

Merging #84 will degrade performances by 44.98%

_{Comparing feat/fpcs (499d538) with main (7e2f14a)}

Summary

⚡ 17 improvements
❌ 6 regressions
✅ 655 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
❌	test_m4_no_x[True-int32-100-100,000]	520 µs	618.6 µs	-15.94%
⚡	test_m4_no_x[True-int64-5,000-100,000]	1,070.9 µs	904.9 µs	+18.34%
❌	test_m4_with_x[True-float64-1,000-100,000]	1.6 ms	2 ms	-20.22%
⚡	test_m4_with_x[True-int32-1,000-1,000,000]	3.8 ms	3.4 ms	+12.05%
❌	test_m4_with_x[True-int32-1,000-100,000]	1.5 ms	1.9 ms	-19.61%
⚡	test_m4_with_x[True-int64-1,000-100,000]	1.9 ms	1.6 ms	+13.28%
⚡	test_minmax_with_x[True-int32-1,000-100,000]	2 ms	1.6 ms	+26.32%
⚡	test_minmax_with_x[True-int64-1,000-100,000]	3 ms	2.2 ms	+33.14%
⚡	test_minmaxlttb_no_x[True-float32-5,000-100,000]	2.7 ms	2.3 ms	+17.68%
⚡	test_nanm4_no_x[True-int32-100-100,000]	645.4 µs	541.7 µs	+19.14%
⚡	test_nanm4_with_x[True-int32-1,000-100,000]	1.6 ms	1.4 ms	+14.21%
⚡	test_nanm4_with_x[True-int64-1,000-100,000]	1.9 ms	1.7 ms	+14.93%
❌	test_nanminmax_with_x[True-float32-1,000-100,000]	1.5 ms	2.7 ms	-44.98%
⚡	test_nanminmax_with_x[True-float64-1,000-1,000,000]	6.5 ms	5.8 ms	+11.66%
⚡	test_nanminmax_with_x[True-float64-1,000-100,000]	2.4 ms	2.1 ms	+11.74%
⚡	test_nanminmax_with_x[True-int32-1,000-1,000,000]	4.1 ms	3.6 ms	+13.56%
❌	test_nanminmax_with_x[True-int32-1,000-100,000]	1.6 ms	2 ms	-21.27%
⚡	test_nanminmax_with_x[True-int64-5,000-100,000]	5.2 ms	4.7 ms	+11.04%
⚡	test_nanminmaxlttb_no_x[True-float32-100-100,000]	610.4 µs	459.3 µs	+32.9%
⚡	test_nanminmaxlttb_no_x[True-float32-5,000-100,000]	2.5 ms	2.3 ms	+10.09%
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.