Introduce an event pair class?

[eyalroz]

CUDA has almost now APIs which regard pairs of events. There's certainly no CUDA driver or runtime struct fort representing event pairs. Consequently, I have no event pair wrapper.

... and yet... event pairs do show up. Timing using events is only defined on pairs of events. Which also motivated #399. In other words, an event pair type has already snuck in to our codebase. So, perhaps we should flesh it out into a full-fledged class?

• Let the user create an event pair rather than creating two events separately
• Offer some kind of wrapped enqueue mechanism for streams, where the two events are used to denote the start and end of the enqueued work
• ... what else?

I have a single decent usage scenario, a second scenario of questionable generality, and... what else? Is this enough for a full-fledged class?