hv2-net::nat: clippy is red again on current stable (chunks_exact_to_as_chunks)

crates/hv2-net/src/nat.rs の internet_checksum から始めます。ここでは data.chunks_exact(2) が報告された lint を発生させます。cargo clippy -p hv2-net --all-targets -- -D warnings で現在の stable toolchain を確認し、lint の処理を stable バージョン間で移植可能にします。lint が存在する toolchain と存在しない toolchain の両方で、このコマンドが成功すれば完了です。

Verifying #95's networking work (all four pieces: device, attach_net, guest driver, bridge) against current stable before treating it as done. Good news first, since it's the more important result:

The full chain works end to end

Ran the net_echo example against /dev/kvm (via WSL) -- the thing the handoff page flagged as "not run": a guest booted, its virtio-net driver came up, and a frame round-tripped through it.

guest         : /tmp/hv2-unikernel/hv2-unikernel.elf
net           : window at 0xd0020000, IRQ 6
guest driver  : up in    7.4 ms
frame sent    : 57 bytes
echo          : 57 bytes back, addresses swapped
round trip    :    2.3 ms
result        : a frame crossed into a guest, through a driver with no allocator under it, and came back

Also independently reran and counted the unit tests the recent commits claim: 9 on virtio_net_mmio, 9 on VM::attach_net, 5 on hv2_net::bridge -- 23, matching. cargo fmt --all -- --check is clean.

But clippy is red on hv2-net right now

cargo clippy -p hv2-net --all-targets -- -D warnings (same command CI runs, per 11997dd's message) fails today:

error: using `chunks_exact` with a constant chunk size
   --> crates\hv2-net\src\nat.rs:418:27
418 |     let mut chunks = data.chunks_exact(2);
    = note: `-D clippy::chunks-exact-to-as-chunks` implied by `-D warnings`

This is the same lint 11997dd removed the #[allow(...)] for, on the grounds that the attribute was itself unknown (unknown_lints) on the toolchain that session had. rust-toolchain.toml pins channel = "stable" -- a floating channel, not a fixed version -- so which lints exist under -D warnings depends on whenever rustup last updated. My local stable right now is 1.98.1 and it does recognize clippy::chunks_exact_to_as_chunks; whatever stable 11997dd's session had did not. Both sessions ran the identical command in good faith and got opposite answers, which is what a floating channel does.

The portable fix -- survives both directions of that drift -- is #[allow(unknown_lints, clippy::chunks_exact_to_as_chunks)] rather than the lint alone: unknown_lints swallows the "this isn't a real lint" complaint on an older toolchain, and the specific allow swallows the real one on a newer toolchain that has it.

#[allow(unknown_lints, clippy::chunks_exact_to_as_chunks)] // may not exist on every `stable`; unknown_lints covers that case too
fn internet_checksum(data: &[u8]) -> u16 {

Worth deciding separately, and out of scope for this issue: whether channel = "stable" should become a pinned version instead, so cargo clippy -D warnings gives the same verdict on every machine and in CI rather than depending on when each one last ran rustup update. This is the second time this exact class of drift has cost a red build over this one function.