review: tighten comments + unsafe SAFETY note

[CMGS]

Whole-repo comment/style pass over sandbox main (/code Go, /code-rs Rust, /code-py Python). The tree was audited file-by-file; it came back already within budget, so this PR is deliberately small.

Changes

• sandboxd/server/server.go — drop the , as before. edit-narration tail from resolveScope's godoc; the sentence is complete without it.
• sdk/openai/.../adapter.py — remove the restated-code comment in _abs; the three-line body (Path coerce → absolute passthrough → join under manifest.root) is self-evident.
• silkd/src/sysutil.rs — add the missing SAFETY invariant (/code-rs rule 5) on lookup_user's non-null passwd dereference: pw is non-null (checked) and, with NSS_LOCK held, points to a valid passwd whose pw_dir is a NUL-terminated string it owns. The other unsafe blocks already carry their invariant (guard comments on kill_group/signal_pid, the NSS_LOCK note on getpwnam, the F_GETFL/F_SETFL note on set_nonblocking, the openpty SAFETY block).

Audit result

Across ~95 source files the pass found only these three items — two removable comments and one genuine unsafe-justification gap. Declaration layout, logging idioms, context propagation, error handling, and English-only were all clean; no manufactured churn.

Gates

• Rust: cargo fmt --check ✓, clippy --all-targets -D warnings on macOS ✓ and in a rust:1-slim Linux container (incl. vsock/tokio-vsock) ✓, cargo test ✓ (10 passed).
• Go: go build/vet ✓, golangci-lint run on GOOS=linux and GOOS=darwin ✓ (0 issues), golangci-lint fmt --diff ✓.
• Python: ruff check ✓, pytest ✓ (4 passed, Python 3.12 venv with both SDK packages editable-installed).