optimize arm64 bitpacking

[achille-roussel]

Based on #5, this change adds specializations for ARM64, with vectorized bit packing for bit widths 1 to 8.

goos: darwin
goarch: arm64
pkg: github.com/parquet-go/bitpack
cpu: Apple M2 Pro
                     │ /tmp/bench_pack_purego.txt │      /tmp/bench_pack_arm64.txt      │
                     │           sec/op           │   sec/op     vs base                │
PackInt32/bitWidth=1                 101.35n ± 1%   26.39n ± 1%  -73.96% (p=0.000 n=10)
PackInt32/bitWidth=2                  90.77n ± 2%   29.83n ± 1%  -67.14% (p=0.000 n=10)
PackInt32/bitWidth=3                  93.82n ± 1%   28.11n ± 1%  -70.04% (p=0.000 n=10)
PackInt32/bitWidth=4                  97.97n ± 2%   30.03n ± 1%  -69.35% (p=0.000 n=10)
PackInt32/bitWidth=8                 112.05n ± 2%   23.12n ± 1%  -79.36% (p=0.000 n=10)
PackInt64/bitWidth=1                 100.80n ± 1%   26.43n ± 2%  -73.78% (p=0.000 n=10)
PackInt64/bitWidth=2                 105.80n ± 3%   29.98n ± 1%  -71.67% (p=0.000 n=10)
PackInt64/bitWidth=3                  95.33n ± 2%   28.22n ± 3%  -70.40% (p=0.000 n=10)
PackInt64/bitWidth=4                  96.04n ± 2%   30.11n ± 3%  -68.65% (p=0.000 n=10)
PackInt64/bitWidth=8                 101.55n ± 1%   23.50n ± 4%  -76.85% (p=0.000 n=10)
geomean                               99.38n        27.45n       -72.38%

                     │ /tmp/bench_pack_purego.txt │       /tmp/bench_pack_arm64.txt        │
                     │            B/s             │      B/s       vs base                 │
PackInt32/bitWidth=1                 4.707Gi ± 1%   18.065Gi ± 1%  +283.81% (p=0.000 n=10)
PackInt32/bitWidth=2                 5.253Gi ± 2%   15.988Gi ± 1%  +204.36% (p=0.000 n=10)
PackInt32/bitWidth=3                 5.082Gi ± 1%   16.964Gi ± 1%  +233.78% (p=0.000 n=10)
PackInt32/bitWidth=4                 4.867Gi ± 2%   15.877Gi ± 1%  +226.23% (p=0.000 n=10)
PackInt32/bitWidth=8                 4.255Gi ± 2%   20.620Gi ± 1%  +384.64% (p=0.000 n=10)
PackInt64/bitWidth=1                 9.461Gi ± 1%   36.090Gi ± 2%  +281.45% (p=0.000 n=10)
PackInt64/bitWidth=2                 9.014Gi ± 3%   31.816Gi ± 1%  +252.97% (p=0.000 n=10)
PackInt64/bitWidth=3                 10.00Gi ± 2%    33.79Gi ± 3%  +237.82% (p=0.000 n=10)
PackInt64/bitWidth=4                 9.930Gi ± 2%   31.670Gi ± 3%  +218.93% (p=0.000 n=10)
PackInt64/bitWidth=8                 9.393Gi ± 1%   40.576Gi ± 4%  +332.00% (p=0.000 n=10)
geomean                              6.786Gi         24.56Gi       +261.99%