Quote gcc manual: "These functions are intended to replace the legacy
‘__sync’ builtins. The main difference is that the memory order that is
requested is a parameter to the functions. New code should always use
the ‘__atomic’ builtins rather than the ‘__sync’ builtins.

Same assembler code generated on x86_64 for both builtins and no matter
what memory order is specified:

    lock xaddl	%eax, debug_remaining_connects(%rip)

On ppc, however, __sync_fetch_and_sub creates code for sequential
consistency and puts a "sync" instruction before the decrement (which is
done atomicially between "lwarx" and "stwcx" instructions) and a "isync"
instruction behind it. These two instructions are omitted when only
relaxed consistency is requested:

        sync           # optional
    L22:
        lwarx 9,0,27
        addi 10,9,-1
        stwcx. 10,0,27
        bne 0,.L22
        isync          # optional

Not, that it would matter at all :-)

Make this change anyway to get used to the recommended builtins.

When SSL sees a client hangup, SSL_get_error currently returns
SSL_ERROR_SYSCALL.

The page at [1] says in section BUGS, that errno would be 0 in this
case:

    The SSL_ERROR_SYSCALL with errno value of 0 indicates unexpected EOF
    from the peer.

To my experimentation, this is not true. errno is left unchanged instead.

The page at [2] says, that ERR_get_error could be used to distinguish
between EOF and some errno indicated failure:

    SSL_ERROR_SYSCALL Some I/O error occurred. The OpenSSL error queue
    may contain more information on the error. If the error queue is
    empty (i.e. ERR_get_error() returns 0), ret can be used to find out
    more about the error: If ret == 0, an EOF was observed that violates
    the protocol. If ret == -1, the underlying BIO reported an I/O error
    (for socket I/O on Unix systems, consult errno for details).

But to my experimentation, this is not true either. In both cases,
ERR_get_error doesn't have any further information and ret (from
SSL_accept) is -1 in both cases.

This might be fixed in OpenSLL 3.0 [3]:

    On an unexpected EOF, versions before OpenSSL 3.0 returned
    SSL_ERROR_SYSCALL, nothing was added to the error stack, and errno was
    0. Since OpenSSL 3.0 the returned error is SSL_ERROR_SSL with a
    meaningful error on the error stack.

Although it is ugly, we only have the option to set errno to zero before
the operation. do this in ssl_read_with_timeout and
ssl_accept_with_timeout.

[1]: https://www.openssl.org/docs/man1.1.1/man3/SSL_get_error.html
[2]: https://linux.die.net/man/3/ssl_get_error
[3]: https://www.openssl.org/docs/manmaster/man3/SSL_get_error.html

Use COMMON_LOG to log failures from wait_rd_with_timeout which includes
a timeout.

The call `psslerror("")` prints a line with ":" only. Make the string
optional so that NULL or "" won't output anything.