= Fixes =
* New config var 'query_wait_timeout'. If client does not get
server connection in this many seconds, it will be killed.
* If no server connection in pool and last connect failed, then
don't put client connections on hold but send error immediately.
This together with previous fix avoids unnecessary stalls if
a database has gone down.
* Track libevent state in sbuf.c to avoid double event_del(). Although
it usually is safe, it does not seem to work 100%. Now we should always
know whether it has been called or not.
* Disable maintenance during SUSPEND. Otherwise with short timeouts
the old bouncer could close few connections after sending them over.
* Apply client_login_timeout to clients waiting for welcome packet
(first server connection). Otherwise they can stay waiting
infinitely, unless there is query_timeout set.
* win32: Add switch -U/-P to -regservice to let user pick account
to run service under. Old automatic choice between Local Service and
Local System was not reliable enough.
* console: Remove \0 from end of text columns. It was hard to notice,
as C clients were fine with it.
* Documentation improvements. (Greg Sabino Mullane)
* Clarify few login-related log messages.
* Change logging level for pooler-sent errors (usually on disconnect) from INFO
to WARNING, as they signify problems.
* Change log message for query_timeout to "query timeout".
