replication sieve activate should be atomic

[brong]

In timsieved it's very careful to always keep a script active:

/* ok we want to do this atomically so let's
   - make <activesieve>.NEW as a hard link
   - rename it to <activesieve>
*/
result = symlink(filename, "defaultbc.NEW");
if (result) {
    syslog(LOG_ERR, "symlink(%s, defaultbc.NEW): %m", filename);
    prot_printf(conn, "NO \"Can't make link\"\r\n");
    return TIMSIEVE_FAIL;
}

result = rename("defaultbc.NEW", "defaultbc");
if (result) {
    unlink("defaultbc.NEW");
    syslog(LOG_ERR, "rename(defaultbc.NEW, defaultbc): %m");
    prot_printf(conn,"NO \"Error renaming\"\r\n");
    return TIMSIEVE_FAIL;
}
sync_log_sieve(sieved_userid);

...

In sync_support, it is not only not atomic, it has a gap while the whole file gets recompiled:

snprintf(target, sizeof(target), "%s", name);
snprintf(active, sizeof(active), "%s/%s", sieve_path, "defaultbc");
unlink(active);

#ifdef USE_SIEVE
char *bc_fname = strconcat(sieve_path, "/", target, NULL);
sieve_rebuild(NULL, bc_fname, 0, NULL);
free(bc_fname);
#endif

/* N.B symlink() does NOT verify target for anything but string validity,
 * so activation of a nonexistent script will report success.
 */
if (symlink(target, active) < 0)
    return(IMAP_IOERROR);

...

sync_support should switch to using the same logic that timsieved does to ensure there's always an active script.