How does co-produced research influence adaptive capacity? Lessons from a cross-case comparison