Streamline CI

[illuzen]

• parallel tests
• download taplo binary instead of building from source
• don't cargo clean before tests
• cache build artifacts with rusty-cache

---

Note

Low Risk
Workflow-only changes; test execution still uses the same locked workspace and feature flags, with only the runner and caching strategy changed.

Overview
CI is tuned for faster feedback by caching Rust builds, compiling through sccache, and running tests with cargo nextest instead of plain cargo test.

The build-and-test and analysis jobs now set SCCACHE_GHA_ENABLED / RUSTC_WRAPPER, add mozilla-actions/sccache-action, and Swatinem/rust-cache with per-job shared-key values (including OS-specific keys on the matrix). The pre-test cargo clean step is removed so the compile step’s artifacts can be reused for tests.

Fast checks install taplo from a released Linux binary via curl rather than cargo install taplo-cli, avoiding a source build on every run.

^{Reviewed by Cursor Bugbot for commit 5f5ca26. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 5f5ca26. Configure here.}

[n13]

Review — Approve

Workflow-only change, and all checks (fast-checks, build & test matrix on ubuntu + macos, analysis) are green on the head commit. The one Bugbot finding (nextest doesn't run doctests) was correctly resolved by the run doctests commit, which adds a dedicated cargo test --doc step. LGTM.

What's good

• sccache + Swatinem/rust-cache + dropping cargo clean should meaningfully cut warm-cache build times.
• cargo nextest run for parallel test execution.
• Downloading the prebuilt taplo binary instead of cargo install taplo-cli --locked avoids a from-source build each run (verified the release asset URL resolves).
• Per-job shared-key values (including OS in the matrix) keep the caches correctly partitioned.

Non-blocking suggestions

1. Pin taplo instead of latest. releases/latest/download/... is non-reproducible — a new taplo release can silently change formatting rules (or asset layout) and break CI without any change on our side. Pin to an explicit version tag to keep formatting deterministic.
2. Watch disk usage from removing cargo clean. The old step was labeled "to free space," and now target/ is also cached. The matrix builds the workspace with runtime-benchmarks,try-runtime (compile) and again with fast-governance (test + doctest), so on the standard hosted runners this could get tight. It passed this run (ubuntu ~3h16m), so fine for now — just flagging it as a fragility to keep an eye on.
3. sccache + rust-cache is somewhat redundant. rust-cache already caches target/ and the registry; sccache adds a separate object cache on top. Harmless, but the marginal benefit may be small — and sccache won't cache the clippy-driver invocations in the analysis job, where rust-cache does the heavy lifting.
4. Branch is named beefier-runners but runs-on is still ubuntu-latest/macos-latest. If bumping to larger runners was part of the intent, it's not in this diff — was that deferred intentionally?

None of these block merge.