v2 Bug fix

This release fixes a bug in temporal denoising that creates visual artifacts. (#2)

benchmark

1. Intel Xeon Platinum 8480+, linux kernel 5.18.15, python 3.9, vapoursynth-classic R57.A8-6-g70a15729, ispc v1.20.0, quadrant
2. Intel Xeon CPU Max 9480, linux kernel 5.15.0, python 3.9, vapoursynth-classic R57.A8-6-g70a15729, ispc v1.20.0, HBM-only, SNC4

data format: fps

1920x1080

Gray16

d	a	8480+	9480
0	2	156.79	400.47
0	3	78.21	201.45
1	2	37.17	86.84
1	3	18.64	43.75
2	2	20.30	38.24
2	3	10.47	23.53
3	2	14.40	31.76
3	3	7.36	16.82

YUV444P16 (channels="YUV")

d	a	8480+	9480
0	2	107.07	295.36
0	3	54.11	149.52
1	2	25.21	65.70
1	3	13.17	32.91
2	2	14.34	35.24
2	3	7.30	17.60
3	2	9.92	23.67
3	3	5.10	12.45

v1 Initial release

This is a cpu-only implementation of KNLMeansCL for VapourSynth.

It has no external dependency and can be at most 10x faster than its counterpart. Performance is severely limited by memory bandwidth, however.