Replace gzip with zstd for speed + size benefits

[jeffwidman]

Replace gzip with zstd for both speed + size benefits.

I added zstd to the base image rather than the python-specific image because:

1. it's only 1,695 KB
2. We'll likely use it in other places down the road, such as unpacking deps for jobs
3. The Python image would have required additional steps like apt-get update
4. Originally I was afraid that installing zstd would take more time than is saved with the compression speedup--and that time would apply to all images, but looking at the raw CI logs and grep'ing for zst shows that it takes < 0.5s to install it.

This results in faster compression + space savings:

root@ca684869d4af:/usr/local/.pyenv/versions# time tar -acf 3.11.5.tar.gz 3.11.5

real	0m5.564s
user	0m5.458s
sys	0m0.626s
root@ca684869d4af:/usr/local/.pyenv/versions# time tar -acf 3.11.5.tar.zst 3.11.5

real	0m0.850s
user	0m1.003s
sys	0m0.354s
root@ca684869d4af:/usr/local/.pyenv/versions# ls -lah
-rw-r--r-- 1 root       root        32M Sep  1 17:19 3.11.5.tar.gz
-rw-r--r-- 1 root       root        30M Sep  1 17:20 3.11.5.tar.zst

As well as faster decompression too:

root@ca684869d4af:/usr/local/.pyenv/versions# ls -lah
-rw-r--r-- 1 root       root        32M Sep  1 17:38 3.11.5.tar.gz
-rw-r--r-- 1 root       root        30M Sep  1 17:36 3.11.5.tar.zst
root@ca684869d4af:/usr/local/.pyenv/versions# time tar -axf $PYENV_ROOT/versions/3.11.5.tar.gz -C $PYENV_ROOT/versions

real	0m1.113s
user	0m0.986s
sys	0m0.811s
root@ca684869d4af:/usr/local/.pyenv/versions# rm -rf 3.11.5
root@ca684869d4af:/usr/local/.pyenv/versions# time tar -axf $PYENV_ROOT/versions/3.11.5.tar.zst -C $PYENV_ROOT/versions

real	0m0.774s
user	0m0.501s
sys	0m0.695s

The -a flag to tar tells it to autoselect the compression format based on the file extension, so it's flipping between gzip and zstd.

[jeffwidman]

I have been meaning for a while to experiment with different compression formats, so took the time to play around with it for a few mins this morning for fun.

@Nishnha this is likely a better solution to the pigz problem in the runner, given how much time is spent in compression, I suspect it'd result in noticeable speed + compression size improvements in that deploy pipeline.