One way to both extend JFR to support arbitrary sampling strategies, and to make it more useful for other use cases too, is to allow stack trace and thread to be filled out async before commit through a supported Java API. Something like,
event.captureThread(thread);

I fully realize how this could complicate the implementation (e.g. emission of constant pools / checkpoint events), but this would make it possible to solve several issues we (Datadog) have today with JFR.