fix: three Linux clippy errors and a macOS test failure; CI has been red

CI has failed on every run going back to at least 11 August — twelve
consecutive, spanning the 5.0.0 and 5.1.0 releases. That is why the `cli`
feature break shipped: the job that exists to catch it was reporting, and a
permanently red pipeline reports nothing anybody reads.

Two causes, neither visible from Windows.

Clippy on Linux: two `map_or(true, ..)` in motherboard/linux.rs and an
`io::Error::new(ErrorKind::Other, ..)` in profile/nvme_features.rs, all in
`cfg(target_os = "linux")` code that local clippy never compiles.

Test on macOS: `revert_without_confirmation_is_refused` asserted
`NeedsConfirm` against `active_scheme_guid`, a Windows-only setting. macOS
registers no writable settings at all, so `apply_setting` refuses at the
handler lookup with `NotWritable` before reaching the confirmation gate.

That ordering is correct and is left alone — telling someone to pass
--confirm for a setting that can never be written on their platform sends
them down a dead end. The test was wrong, not the code: it asserted a
Windows fact on every platform, passed on the machine that wrote it, and
failed everywhere else.

It now asserts the invariant that does hold everywhere — an unconfirmed
revert never writes — and then the platform-appropriate refusal, asking the
handler registry which case applies rather than restating its cfgs, so a
handler that gains or loses a platform cannot make the test vacuous.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>