Use np.packbits in half int compression

[jonashaag]

No description provided.

[github-actions]

(benchmark 4461990141 / attempt 1)
Base results / Our results / Change

Model	Size	Dump Time	Load Time
sklearn rf	77.2 MiB / 11.1 MiB / 6.94 x	0.07 s / 0.11 s / 1.49 x	0.08 s / 0.10 s / 1.25 x
sklearn rf LZMA	17.3 MiB / 5.3 MiB / 3.24 x	34.75 s / 5.55 s / 0.16 x	1.63 s / 0.57 s / 0.35 x
sklearn gb	2.2 MiB / 1.1 MiB / 2.08 x	0.03 s / 0.17 s / 5.83 x	0.03 s / 0.13 s / 3.89 x
sklearn gb LZMA	0.6 MiB / 0.2 MiB / 3.82 x	0.95 s / 0.36 s / 0.38 x	0.09 s / 0.14 s / 1.61 x
LGBM gbdt	5.3 MiB / 1.9 MiB / 2.81 x	0.16 s / 0.39 s / 2.42 x	0.02 s / 0.31 s / 12.37 x
LGBM gbdt LZMA	1.7 MiB / 0.8 MiB / 1.96 x	3.05 s / 0.97 s / 0.32 x	0.16 s / 0.38 s / 2.34 x
LGBM gbdt large	5.3 MiB / 1.9 MiB / 2.81 x	0.17 s / 0.40 s / 2.34 x	0.02 s / 0.31 s / 12.51 x
LGBM gbdt large LZMA	1.7 MiB / 0.8 MiB / 1.96 x	3.05 s / 0.97 s / 0.32 x	0.16 s / 0.38 s / 2.35 x
LGBM rf	1.6 MiB / 0.5 MiB / 3.42 x	0.05 s / 0.08 s / 1.84 x	0.01 s / 0.08 s / 13.09 x
LGBM rf LZMA	0.1 MiB / 0.1 MiB / 1.70 x	0.23 s / 0.12 s / 0.52 x	0.02 s / 0.08 s / 4.24 x

[pavelzw]

RF
6.63x -> 6.94 x

RF LZMA
3.23x -> 3.24x

GB
2.06x -> 2.08x

GB LZMA
3.82x -> 3.82x