Use classed conditions as often as possible

[MichaelChirico]

#5912 is the latest example, but generally speaking it is a good idea to create some classed conditions.

The first step is to add support to stopf() and friends for making the resulting condition with extended class.

Thereafter we can work towards a taxonomy of condition classes.

[MichaelChirico]

We'll also benefit from putting ... into the signature of test(), with entries referring to condition classes, then we can do test(n,x,y, custom_warning = ...). Implementation gets a bit messier from there but should be doable.

[jangorecki]

I would also wait with this one till 1.17.0 at least, when first round of pending PRs will get resolved

[MichaelChirico]

WDYT about merging #5914 now?

We don't need to use classed conditions everywhere yet, but could at least start for some common errors. Changing stopf() (etc) lines shouldn't cause too many conflicts with other PRs.

Even if not, with #5914 we can start writing custom classes for new functionality.

Most important is to pick a framework for errors, i.e. should all errors be classed like c("dt_${SPECIFIC_CLASS}", "dt_warn", "warning") (similarly for dt_error, dt_message, dt_startup_message)? Something else? We'll want to pick that first to avoid needing to change it when we scale up eventually.