Re-introduce support for bcrypt

[wohali]

In preparing the 2.2.0 release, it was found that the bcrypt add-on will not compile under Windows. There are at least 2 reasons for this:

1. Windows does not include the BSD libc extensions, meaning there is no sys/queue.h.
2. Our Windows build environment, which does not use msys/mingw/etc, does not have a useful pthread library.

In addition, it seems wrong to me that a NIF to just run an encryption routine is launching its own threads. Threading and scheduling of those threads should be left up to the Erlang scheduler. This approach, which was borrowed from esnappy, was appropriate when it was written (R14 era), but with improved (dirty) schedulers in Erlang 17+, it's no longer the best way. Perhaps in removing this code for Windows, we can improve the implementation for other platforms at the same time.

Because of the level of integration of bcrypt, it is not feasible to remove the library on one platform, or at configure time. An attempt to do that by @iilyak was aborted, see #1228.

As such, the bcrypt application is being removed from master.

Expected Behavior

bcrypt functionality should be restored to CouchDB post-2.2.0. That reworked implementation should work under Windows, and should not rely on pthreads.

Current Behavior

Windows build is broken, full stop.

[wohali]

The bcrypt branch has been restored, and has cherry-picked commits on it for all of the merged PRs that improved its test suite.

Feel free to use this for work going forward, or if a rewrite is needed, reimplement from scratch.

[nickva]

PBKDF2 implementation was just improved in #4814

https://csrc.nist.gov/csrc/media/Projects/crypto-publication-review-project/documents/initial-comments/sp800-132-initial-public-comments-2023.pdf indicates Argon2 looks like a favorite of industry reviewers so far instead of brcypt. Since we haven't brought bcrypt back in 5 years, it's probably worth closing the issue.