perf(nix): reuse musl artifacts and remove redundant release checks

[davidabram]

Summary

• Pass prebuilt musl artifacts to Crane through its expected cargoArtifacts argument. The old inherit cargoArtifactsMusl was ignored, causing Crane to create another dependency derivation on Linux.
• Use cargoWithProfile build --locked in the native and musl release dependency derivations, skipping the redundant cargo check --release pass. Test and Clippy derivations stay unchanged.

Motivation / baseline

• Parent: select multiple integration targets in interactive setup #321 (cool-stuff); this PR is stacked directly on top of it.
• PR select multiple integration targets in interactive setup #321 Linux release validation took 17m 58s. Logs show sce-deps-musl-deps (~6m 50s), then a separate implicit sce-deps (~6m 47s), then the final binary (~58s).
• PR select multiple integration targets in interactive setup #321 macOS release dependency phase took ~8m 37s, including a cargo check --release pass (~3m 30s).
• Those are observations, not claimed projected savings. Measure the actual CI outcomes against select multiple integration targets in interactive setup #321 and fix(build): avoid invalidating Cargo on generated OUT_DIR inputs #318; account for GitHub Actions / Magic Nix Cache variability.

Validation

• nix flake check --fallback --print-build-logs on Linux/macOS
• nix build .#ci-checks --fallback --print-build-logs on Linux/macOS
• Linux musl release portability audit passes
• Both release binaries still embed the correct commit and run sce --help / sce version as currently covered by CI
• Linux release logs no longer build the implicit second sce-deps derivation; compare elapsed times

Scope: flake.nix only. No changes to release targets, checks, or toolchain.