What really matters for successful research environments? A realist synthesis