test(detach): ask the marker rule, not the machine the suite runs on

[whackur]

cargo test fails on a_process_with_no_marker_is_the_foreground_copy
when the suite is run from inside a nightcrow pane. Not a flake and not
a Windows quirk: the pane's shell inherits NIGHTCROW_DETACHED from the
daemon that spawned it, and the test read that same variable through
is_detached_child().

thread 'daemon::detach::tests::a_process_with_no_marker_is_the_foreground_copy'
panicked at src\daemon\detach_tests.rs:15:5:
assertion failed: !is_detached_child()

So the test asserted a fact about the machine it ran on rather than
about the rule it names. CI never saw it, because CI is not inside a
nightcrow pane — but anyone dogfooding the tool hits it on every run,
including through the pre-push gate.

The change

is_detached_child read the environment and decided in one step. Those
are now two steps, the way child_args and background_command are
already split from spawning:

pub fn is_detached_child() -> bool {
    marker_says_detached(std::env::var_os(ALREADY_DETACHED).as_deref())
}

fn marker_says_detached(marker: Option<&OsStr>) -> bool {
    marker.is_some()
}

The tests pass the marker in. Both answers are stated outright instead
of one of them depending on ambient state, and a third case the rule
quietly carried is pinned: presence is what counts, so an empty value is
still a marker and must not send the child off to detach again.

Verification

Full suite green with NIGHTCROW_DETACHED=1 still set in the shell —
1419 passed, 0 failed — which is the case that used to fail. clippy -D warnings and fmt --check pass; the pre-push gate now runs clean from
inside a pane without clearing the variable first.