Parallelize ObjectStorageProvider downloads

[petamas]

After starting to use Boost from vcpkg in our project, the number of used ports jumped from 20 to ~100. Similarly, restoring all packages from an AWS S3 backend jumped from 2 minutes to almos 10 minutes, which is way too long for us.

Since AWS has a rate limit on individual file downloads, but there's no limit on downloading files in parallel, I've decided to parallelize the download process on the ObjectStorageProvider level.

I hope this doesn't break Google Cloud Storage and Tencent Cloud Object Storage backends, but I have no way of testing them. If needed, I can add a supports_parallel_downloads() method to IObjectStorageTool, and constrain the parallelization to AWS.

[autoantwort]

I guess this is more or less the same as in #1392?

[petamas]

You're right, it's basically the same. I've also experienced about 5x speedup using this parallelization, similar to @Crzyrndm.

However, if this comment is right, i.e. printing is not thread-safe, then my solution has the same issue. Unfortunately, the fix is not trivial, because extracting out_sink.println_warning() is not enough: the underlying cmd_execute_and_capture_output may also print in case --debug is passed. :-/