Optimising participation and generalisability: the use of opt-out recruitment for an implementation trial in primary care