perf(nix): speed up Rust checks and deduplicate Quint MBT

[davidabram]

Pull request stack

Review and merge bottom-up. Each PR targets the previous PR's branch, so Files changed shows only that step's incremental changes.

Step	Pull request	Change
1/4	#308 — Faster Nix checks	Rust test-profile cache; deduplicate default MBT check
2/4	#309 — Quint workflow cleanup	Stop recompiling Rust MBT separately
3/4	#310 — macOS smoke-build reuse	Reuse release artifact for macOS smoke tests
4/4	#311 — CI timing summaries	Record Nix phase durations in job summaries

You are reviewing step 1/4: #308. Parent: main. Next: #309.

---

Summary

• Compile Rust tests and Clippy with Crane's test profile instead of the default optimized release profile, using line-table-only debug info.
• Introduce cargoArtifactsTest and pass it consistently to both cli-tests and cli-clippy so they reuse a matching dependency cache. The existing release-profile cargoArtifacts and distributable CLI builds are unchanged.
• Remove the separately compiled/running mutation-trace-quint-connect check from the default flake checks set. The full cli-tests suite already runs those 16 MBT tests.
• Expose the focused Quint MBT runner as the opt-in packages.mutation-trace-quint-connect and update .github/workflows/quint.yml to call its new attribute. Preserve its pinned Rust and Quint environment, and document both execution modes.

Motivation

The October 9 baseline Nix CI jobs showed nix flake check taking 20m34s on Linux and 28m59s on macOS. Most wall time was Rust dependency/test compilation, not Nix evaluation. Prior to this patch, cli-tests and the separate Quint MBT check each compiled a release-profile test binary and ran the overlapping MBT suite.

CI timing comparison

Actual wall-clock duration of time nix flake check --fallback --print-build-logs from successful GitHub Actions jobs on October 9, 2026 (lower is better):

Platform	Before	PR #308	Saved	Faster
Linux	20m 34.101s	9m 21.892s	11m 12.209s	54.5%
macOS	28m 59.715s	15m 02.847s	13m 56.868s	48.1%

• Before: Nix CI #37967537493, Linux job 113945871110, macOS job 113945871153.
• After: Nix CI #37978379400, Linux job 113982336224, macOS job 113982335618.
• Both the Nix CI and Quint workflows completed successfully.
• New test-profile dependency artifacts compile as intended. Linux: 1,753 tests passed; macOS: 1,749 tests passed; no failures on either host. No extra MBT derivation appears in the default flake-check build.
• Important trade-off: the separate nix build .#sce step took 6m 02.125s (Linux) / 9m 04.637s (macOS) afterward, versus 44.083s / 42.341s before, because the release-profile dependency build is now deferred until the native CLI build step. End-to-end job time improves less than the headline flake-check metric.
• Both runs reported HTTP 418 / rate-limit errors from Magic Nix Cache, so these are measured outcomes under the available CI cache conditions, not a controlled warm-cache benchmark.

Validation

When CI runs, verify:

nix flake check --no-build
nix build .#checks.x86_64-linux.cli-tests -L
nix build .#checks.x86_64-linux.cli-clippy -L
nix flake check -L
nix build .#mutation-trace-quint-connect -L
nix build .#sce -L

Confirm that cli-tests includes all MBT tests once, and the focused package executes the same 16 tests on explicit invocation. Check Cargo logs for --profile test in test/Clippy derivations and --profile release for native CLI builds.

Status: GitHub Actions Nix CI completed successfully on Linux and macOS; Quint, CodeQL, and Dependency Review also passed. This environment itself has no Nix installation, so results above are measured from GitHub Actions logs.

Trade-off: Test- and release-profile dependency artifacts are necessarily separate. This improves recompilation of tests while potentially adding a second dependency build on completely cold CI stores. Benchmark end-to-end cold and warm builds before asserting a performance gain.

Scope

Four files: flake.nix, Quint CI workflow, existing MBT context documentation, and existing flake-performance context documentation. No changes to Rust application code, tests, Rust toolchain pins, or production build logic.