- We can reuse the tarball and append new files (node binaries in this
case) instead of compressing/extracting multiple times. This prevents
un-necessary copying of files in the built tarball.

We can also build all `targets` in parallel in the following way:
- Download Node.js versions for targets in parallel instead of serially
- Use [workerpool][1] to create gzip files in parallel
- Run all `xz` compression with maximum available threads. xz has
[supported multicore computation since 5.2 (2014)][1].
- Like gzip, run all `xz` compression in parallel instead of waiting for
each target to build.

Running this in the Heroku CLI produced identical tarballs in 65% amount
of time (~111s seconds vs ~344 seconds before this change). It's likely
even a bit faster since I was running it locally which has some boot-up
time due to ts-node compiling typescript at runtime.

[1]: https://www.npmjs.com/package/workerpool
[2]: https://www.phoronix.com/scan.php?page=news_item&px=MTg3MDM