[narfbg]

Similar to GH-24302 but for Argon2.

ext/password:

• password_hash(), password_needs_rehash() no longer accept invalid $options['threads'] (above 255)
• password_verify() no longer calls libargon2 on malformed hashes
• password_get_info() no longer parses malformed hashes
• Cleanups all around, including fewer heap allocations, but didn't touch the salt handling because that's pending in Make password salt generation algo-agnostic and isolate bcrypt-specific logic #24281. A couple more nitpicky things can be improved as well, but didn't want to balloon the patch even more upfront.

ext/openssl:

• [openssl_]password_hash(), [openssl_]password_needs_rehash() no longer accept invalid $options['threads'] values (above 255)
• [openssl_]password_verify() no longer triggers computations and/or throws exceptions on hashes with invalid parameters
• [openssl_]password_verify() no longer accepts obviously invalid hashes such as with invalid versions or empty salt
• [openssl_]password_hash(), [openssl_]password_needs_rehash() no longer throw exceptions for invalid options, to match ext/password
• [openssl_]password_hash(), [openssl_]password_verify() no longer confuse the threads option with OpenSSL's internal threads (this might be why --with-openssl-argon2 --enable-zts builds are currently forbidden, but haven't tested)

Also (for both) password_needs_rehash() will return true if it sees an older version hash, not just changed parameters. And tests, lots of tests. 🙂

Most of the work I did in ext/password, and tried to keep ext/openssl changes minimal (sscanf() usage still broken), but I do intend to fully fix it up once some decisions are made - whether the same parser approach is ok, exactly which conditions should warrant a rehash, etc.
libsodium provider is mostly fine - nothing really broken; only a bit rough around the edges with error messages and obscure but valid Argon2 v.1.0 hashes. Will polish it later probably alongside OpenSSL's remaining issues.