Document side effects of := on keys

[ben-schwen]

The use of copy() here is subtle, it might be good to describe how to approach side effects a bit better in the documentation.

Originally posted by @MichaelChirico in #7401 (comment)

This should be added to our reference semantics vignette and probably to help(':=') if its not already there.

[MichaelChirico]

Oh, in the PR, what I meant is that it looks like

test(n, optimization=c(1, 2), copy(DT)[...], ...)

Will run copy() for each level of optimization, so there's a lazy evaluation aspect to x when optimization= is provided that impacts how to read and write tests using the new argument. Does that make sense?

[ben-schwen]

I see. In principle it is that seem as for other tests. If multiple execution of a single test works, then the execution of different optimization levels will also work. If a tests already has side effects such that running it twice will error, then ofc running it with different optimization levels will also error.

Hence, one could say that

test(n, optimization=c(1, 2), ...)

is short for

test(n, options(datatable.optimize=1), ...)
test(n+eps, options(datatable.optimize=2), ...)