fix(lint): replace drain(..).collect() with mem::take for clippy 1.98

[cspinetta]

Description

main is currently red on Lint (ubuntu-latest) and Lint (ubuntu-24.04-arm), and every open pull request fails the same way. Nothing in the code changed — the toolchain did.

GitHub's runners moved from rustc 1.97.1 to 1.98.0 (released 2026-08-18) on 2026-08-20. Rust 1.98 promoted clippy::drain_collect, and the lint job runs -D warnings against whatever stable resolves to, so three pre-existing sites began failing on the next run. main's last green CI run was 2026-08-19, on 1.97.1.

Type of Change

• Bug fix (non-breaking change which fixes an issue)

Changes Made

Three drain(..).collect() calls become std::mem::take:

• src/devices/vsock_connection.rs — twice, handing a connection's queued guest-bound bytes to the caller
• src/network/slirp.rs — once, taking the pending DNS queries before resolving them

Every site drains a Vec<T> into a Vec<T>. mem::take swaps in an empty vector and returns the original, which is observably identical — the source ends up empty either way — and skips the reallocation collect performs to rebuild a vector the source already held.

Testing

• All existing tests pass (cargo test --workspace)

Test Commands Run

Verified on a Linux host after updating it to the same toolchain the runners now use, since neither of my machines could reproduce otherwise (one was on 1.97.1, the other on 1.94):

rustup update stable          # 1.97.1 -> 1.98.0 (88d9e12ae 2026-08-18)

# main, to confirm the breakage is pre-existing and not from this branch:
git checkout origin/main
cargo clippy --workspace --all-targets --all-features -- -D warnings
# → 3 errors: draining all elements of a collection into a new collection of the same type

# this branch:
cargo clippy --workspace --all-targets --all-features -- -D warnings   # clean
cargo test --workspace --all-features                                  # pass

Local validation

• Host: Fedora x86_64 (dev1), rustc 1.98.0 — matching the runners
• VM suites: not run (no VM, device, or runtime behavior changes; three expression-level rewrites with identical semantics)
• Perf bench: not run. mem::take removes an allocation and a copy on each of these paths, so the change can only help; measuring it is not worth a bench run.
• Security review: none (no behavior change)
• Perf review: none

Code Quality

• Code follows project style guidelines (cargo fmt)
• No clippy warnings (cargo clippy --workspace --all-targets)

Additional Notes

This will recur. dtolnay/rust-toolchain@stable floats, so any Rust release can promote a lint that reddens every pull request at once, for code nobody touched — and it lands as a required-check failure rather than as a signal that a toolchain moved. Worth deciding separately whether the lint job should pin a version and bump it deliberately; I have not changed that here, since it is a policy question rather than part of unbreaking the build.

A second, smaller gap this exposed: contributors' machines can sit several releases behind the runners, so a local cargo clippy proves less than it appears to. Mine was on 1.94 (March) and the Linux validation host on 1.97.1; neither could reproduce a failure the gate had already hit.