Suggestion: Better ergonomics around collecting multiple kernels

Hey there, I'm really enjoying using KernelAbstractions so far!

Doing Multievents feels clunky, from looking at the source I've used it as follows:

events = []
for layer in 1:layers
    push!(events, kernel_func!(cuda_grids, layer))
end
wait(MultiEvent(Tuple(events)))

Perhaps I missed something, but a suggestion would be to make wait accept an Array{Event,1}?
Under the hood, I understand it's using a tuple. Is there any limitation to the way that wait is implemented such that it couldn't accept an Array?

For instance:

events = [box_blur!(model.grids, layer) for layer in 1:length(model.agent_configs)]
wait(events)

Thanks,
Tom

Start by locating the MultiEvent and wait entry points mentioned in the issue, then inspect how event collections are currently handled. Compare the existing tuple-based behavior with the proposed Array{Event,1} and comprehension examples; done means the requested collection form works without breaking the current usage.