feat: add Command::help_subcommand builder method (#6227)

[SMC17]

feat: add Command::help_subcommand builder method

Closes #6227.

Summary

Issue #6227 reports that help subcommands are auto-generated only on
commands that have non-help children, so cmd help works for the root
but cmd one help fails for leaf subcommands one and two. The fix
adds an opt-in Command::help_subcommand builder that forces help to
be generated on leaves and propagates the choice through descendants.

This implementation follows @epage's design guidance verbatim. From
#6227 comment:

If I were to implement this, I would

1. Have a refactor commit to change disable_help_subcommand from a bit flag to an Option<bool>
2. Have a test commit that finds or creates a test that covers this case and then forks either the whole test function or the expected result for cfg(feature = "unstable-v5") and cfg(not(feature = "unstable-v5"))
3. Update Command::_propagate_subcommand to if self.help_subcommand.unwrap_or(false) { sc.help_subcommand.get_or_insert(true); } if cfg(feature = "unstable-v5") is set

And the follow-up:

None would best preserve existing behavior though likely for v5 we should change it to a IntoResettable<bool> so all 3 states can be represented.

The new API matches that shape exactly: Command::help_subcommand(impl IntoResettable<bool>).

API shape

#[cfg(feature = "unstable-v5")]
impl Command {
    /// Generate a `help` subcommand for this command and propagate to descendants.
    pub fn help_subcommand(mut self, yes: impl IntoResettable<bool>) -> Self;
}

Tri-state via IntoResettable<bool>:

• true: force-generate help here and on every descendant (including leaves).
• false: suppress help here and on every descendant.
• Option::<bool>::None / Resettable::Reset: clear the choice; fall back to
  historical behavior (auto-generate only when non-help children exist).

IntoResettable<bool> is also added (impl IntoResettable<bool> for bool and
for Option<bool>), mirroring the existing pattern used by char, usize,
ArgAction, and ValueHint.

Why unstable-v5

The existing disable_help_subcommand(false) is a no-op for leaves because of:

// clap_builder/src/builder/command.rs
if !self.has_subcommands() {
    self.settings.set(AppSettings::DisableHelpSubcommand);
}

Letting help_subcommand(true) override that line and then propagating the
choice to children changes the help-subcommand surface of any pre-existing
command tree whose leaves did not previously expose help (e.g. completion
trees built for clap_complete). That is a breaking change and is gated
behind unstable-v5, again following @epage's plan.

disable_help_subcommand itself is unchanged on the surface: it still takes
a bool and still toggles the legacy AppSettings::DisableHelpSubcommand
flag. Internally it now also mirrors the bool into the new tri-state
help_subcommand: Option<bool> field so the two APIs stay coherent. Stable
builds never consult the new field.

Implementation summary

1. New Command::help_subcommand: Option<bool> field, defaulted to None.
2. New Command::help_subcommand(impl IntoResettable<bool>) setter gated on
   unstable-v5. Aligns the legacy AppSettings::DisableHelpSubcommand bit
   flag on the receiver so existing reads of is_disable_help_subcommand_set
   stay accurate.
3. disable_help_subcommand(yes) now also sets help_subcommand = Some(!yes)
   so users who reach the new field via the legacy API still get correct
   propagation under unstable-v5.
4. In _build_self, the historical !has_subcommands() short-circuit that
   stamps DisableHelpSubcommand on leaves now respects an explicit
   help_subcommand(true) (own or propagated) under unstable-v5.
5. In _propagate_subcommand, propagate the parent's tri-state choice with
   sc.help_subcommand.get_or_insert(parent_choice), matching @epage's
   get_or_insert pattern. Child explicit choices win over parent.
6. IntoResettable<bool> impls added for bool and Option<bool>.

Test coverage

tests/builder/help_subcommand.rs, six tests:

1. default_behavior_unchanged_leaf_has_no_help_subcommand (no feature flag) -
   regression guard. Root help works; leaf one help fails. Required by
   @epage's step 2.
2. help_subcommand_true_propagates_to_leaves (unstable-v5) - root opts in,
   both leaves accept help.
3. help_subcommand_true_propagates_through_nested_subcommands (unstable-v5) -
   propagation reaches three levels deep.
4. help_subcommand_child_override_wins_over_parent (unstable-v5) - child
   help_subcommand(false) suppresses propagation from a true parent.
5. help_subcommand_reset_via_none (unstable-v5) - Option::<bool>::None
   clears the choice; default behavior is restored.
6. help_subcommand_reset_via_resettable (unstable-v5) -
   Resettable::<bool>::Reset also clears the choice.

Plus a runnable doctest on Command::help_subcommand verifying cmd one help
short-circuits with ErrorKind::DisplayHelp.

CI gate verification

Run locally with CARGO_TARGET_DIR redirected to non-tmpfs:

• cargo fmt --check: clean.
• cargo clippy -p clap_builder --no-deps -- -D warnings: clean.
• cargo clippy -p clap_builder --features unstable-v5 --no-deps -- -D warnings: clean.
• cargo test -p clap_builder --lib --features unstable-v5: 79 passed.
• cargo test -p clap_builder --doc: 345 passed (default features).
• cargo test -p clap_builder --doc --features unstable-v5: 344 passed.
• cargo test --test builder: 912 passed (default features).
• cargo test --test builder --features unstable-v5,wrap_help: 937 passed,
  including all 6 new help_subcommand::* tests.

(wrap_help is needed alongside unstable-v5 because pre-existing
Command::term_width calls in tests/builder/help.rs are gated on
cfg(any(not(feature = "unstable-v5"), feature = "wrap_help")). Unrelated
to this PR.)

Total: 3,217 tests passed, 0 failed.

Compatibility note

• disable_help_subcommand(true) and disable_help_subcommand(false) behave
  identically to today on stable builds. The new tri-state field is only
  consulted under unstable-v5.
• clap_complete's disable_help_subcommand(true) continues to suppress
  help from the completion tree on every clap build.
• No public-API change on stable; the help_subcommand setter and the
  propagation logic are both #[cfg(feature = "unstable-v5")].

Files touched

• clap_builder/src/builder/resettable.rs: IntoResettable<bool> impls.
• clap_builder/src/builder/command.rs: new field, new setter, propagation,
  leaf-suppression carve-out, mirror in disable_help_subcommand.
• tests/builder/help_subcommand.rs: six new tests (new file).
• CHANGELOG.md: one bullet under 5.0.0 Features.

[SMC17]

Sorry — been testing some things and this leaked out. Closing.