I don't see a way to make the testing any more thorough, as we're
basically testing the platform's support for transliteration — there is
nothing our tests can do to enable or disable that support.

They all pass, even with a dummy implementation that just returns
`tocode`. I'll ameliorate that in the next commit.

This test currently fails, because Rust has a dummy implementation.
As a reminder: we can't rely on conventional tests here because
translit() is platform-independent, and there's no fixed known-good
output for any given input. But we can compare two implementations to
see if they behave the same — which is exactly what the test is doing.

The "comparison test" which compares results of C++ and Rust
implementations now pass, and will be removed in the next commit.

This commit passed the CI, i.e. it works not just on my Linux machine,
but also on Ubuntu Linux with various compilers, two macOS versions, and
two FreeBSD versions.

The inner call to iconv_open() creates a separate iconv_t value which is
never deallocated. The impact of this is minuscule, since Newsboat runs
through this code path at most once, but it's still nice to plug this
little leak.

Unlike C++ tests, Rust tests don't show output if they pass. As
a result, the user won't see warnings about the tests that are skipped
due to missing locales. The next commit will fix that.

As mentioned in the previous commit, Rust doesn't display stdout of
tests which passed. As a result, we can't print a warning if we
conditionally skip a test.

An interested user can run `cargo test -- --show-output` and see if
there are any warnings. But this is even worse than our C++ approach: in
C++ it's sufficient to *look* at the output, but with Rust we'd have to
*explicitly ask* for that output.

So this commit moves the decision up, to the *invocation of Make*: `make
check` (and `make ci-check`) runs all the tests *except* the ones that
need locales, while `make NEWSBOAT_RUN_IGNORED_TESTS=1 check` runs
*everything* (ditto for `ci-check`). This way, users and package
builders can continue to run `make check` even if they don't have all
the locales, and our CI can run a more strict version that checks
everything.

It's not a user-facing documentation.

Most of it is (rightfully) marked as outdated, but the section on tests
explains the purpose of `NEWSBOAT_RUN_IGNORED_TESTS` variable that
I introduced a few commits ago.