Update test for unwind::protect error in vroom_errors::warn_for_errors()

[sbearrows]

Fixes #445

This might be a bit redundant/circular, but I'm going to relink @DavisVaughan issue post about this because I found it really useful for refreshing my memory on the problem

r-lib/cpp11#274

[sbearrows]

@jennybc This is ready for you to review

[jennybc]

The update to the test looks good.

I'm pre-approving, but am also suggesting a change to this load-bearing comment, which is a key piece of documentation for something quite subtle.

[jennybc]

vroom::warn_for_errors() warns with base R's machinery rather than cpp11's

This review prompted me to read 👆 for the first time in a long time and then compare it to the source for warn_for_error():

vroom/src/vroom_errors.h

Lines 85 to 103 in f21f314

	void warn_for_errors() const {
	if (!have_warned_ && rows_.size() > 0) {
	have_warned_ = true;
	// it is intentional that we aren't using cpp11::package
	// https://github.com/tidyverse/vroom/commit/984a3e5e37e124feacfec3d184dbeb02eb1145c4
	static auto cli_warn = Rf_findFun(
	Rf_install("cli_warn"),
	Rf_findVarInFrame(R_NamespaceRegistry, Rf_install("cli")));
	cpp11::strings bullets({
	"w"_nm = "One or more parsing issues, call {.fun problems} on your data frame for details, e.g.:",
	" "_nm = "dat <- vroom(...)",
	" "_nm = "problems(dat)"});
	cpp11::sexp cli_warn_call = Rf_lang3(
	cli_warn,
	bullets,
	Rf_mkString("vroom_parse_issue"));
	Rf_eval(cli_warn_call, R_EmptyEnv);
	}
	}

With fresh eyes, I feel like we can refine this to be even more precise. I'd say we're warning with cli and we're using base R machinery to access the relevant R function, as opposed to accessing it via cpp11 way.