Use zstandard (zstd) compression for intermediate artifacts

Coin currently uses parallel gzip for compression. Currently we have builds that produce even 100GB of artifacts, that take more than 5min to compress and timeout. As a result, we resort to not-so-nice hacks, like building the tests in the "Test" workitem, to avoid transfer of artifacts.

Zstd has several advantages, among which:

• Always very fast decompression
• Very configurable compression
  • in the default setting -3 it is much faster than gzip and ratio marginally better.
• Has many useful tweaks for all speed/ratio needs.
  • For example --long increases the compression ratio a lot, together with the memory requirements.
• Supports multiple threads natively (just pass -T0 to the command line).
• Auto-verifies integrity while decompressing, so no extra verification step is needed, like we currently do gzip -t.
• Binaries can be found for almost every OS.

I've discussed it extensively with tosaario and there are a couple of potential drawbacks:

• Integration in golang is dubious.
  • However zstd comes as a mature gzip-like utility and very optimized C library. It might be worth piping the data from golang to the zstd utility while compressing/decompressing. Or just spawning a shell pipeline for the full job, for example curl http://... | zstd -d | tar -xf - for streaming-decompressing-extracting.