Fix: controller: handle remote connection start timeouts correctly

16346686 (2.1.2-rc1) introduced a regression in the handling of
remote connection start timeouts.

It simplified setting failures for handle_remote_ra_start(), but overlooked the
case in retry_start_cmd_cb() where handle_remote_ra_start() is not called
because there is not enough time remaining.

Before that commit, the timeout would get PCMK_OCF_UNKNOWN_ERROR and
PCMK_EXEC_ERROR as the result. Now, fix the regression, and change that to
PCMK_EXEC_TIMEOUT with a suitable exit reason.