wc: match GNU's error for a bad --total value

[arbelonson-source]

wc --total=bogus reports clap's wording rather than GNU's.

$ wc --total=bogus f            # ours, before
error: invalid value 'bogus' for '--total <WHEN>'
For more information, try '--help'.

$ wc --total=bogus f            # GNU 9.11
wc: invalid argument 'bogus' for '--total'
Valid arguments are:
  - 'auto'
  - 'always'
  - 'only'
  - 'never'
Try 'wc --help' for more information.

GNU also distinguishes a value that names nothing from one that names too much — --total=a abbreviates both auto and always, so it is ambiguous, not invalid, and an empty --total= abbreviates all four:

$ wc --total=a f
wc: ambiguous argument 'a' for '--total'
Valid arguments are:
...

And GNU checks every --total, not only the one that wins: wc --total=bogus --total=only f is an error even though the bad value is overridden.

What changed

• The value is validated in Settings::new rather than by ShortcutValueParser, so the message is ours to write. Two new WcError variants carry it.
• TotalWhen::parse keeps the abbreviation handling ShortcutValueParser already gave us: --total=o is only, --total=al is always, --total=a names two choices and so names neither. This also removes the unreachable!("Should have been caught by clap") arm that the old From<T> impl needed.
• The option uses ArgAction::Append so an overridden value is still there to be checked; the last one still decides.
• --help is byte-identical: the option already had hide_possible_values(true), so dropping the value parser changes nothing visible there.

Same shape as the numfmt --round fix in #14292.

Not covered

An invalid-UTF-8 value still gets clap's error: invalid UTF-8 was detected in one or more arguments where GNU says wc: invalid argument '\377' for '--total'. That is unchanged by this PR — main behaves identically — and fixing it means taking the option as an OsString, which felt like a separate change.

Testing

• 75-case differential sweep against GNU 9.11 under LC_ALL=C — stdout, stderr and exit code all match: every valid value, the abbreviations (au, al, n, ne, nev, o, on, onl), near-misses (all, alw, onlyx), wrong case (ALWAYS, Only), values with stray whitespace, the empty value, all 25 --total=X --total=Y permutations, --tot=/--to= prefixes, and combinations with -l, -cmwL and two operands.
• 4 new tests in test_wc.rs, each verified to fail without the change.
• cargo test --features wc --test tests -- test_wc: 63 passed, 0 failed (59 pre-existing, 4 new)
• cargo fmt --check and cargo clippy -p uu_wc --all-targets: clean

Disclosure

Prepared with AI assistance (Claude Opus 5, via Claude Code), per the AI policy in CONTRIBUTING.md. The GNU messages were established by running the installed GNU binary as a black box; I did not read GNU coreutils source. All testing was run locally.

[github-actions]

GNU testsuite comparison:

Skipping an intermittent issue tests/cut/bounded-memory (passes in this run but fails in the 'main' branch)
Congrats! The gnu test tests/tail/tail-n0f is now passing!

[arbelonson-source]

Closing proactively, same reason as #14292 — pure wording alignment over clap's own already-correct validation for --total, no functional change. See the discussion on #14308.