ci: Phase 1 low-risk CI wall-time optimizations

[IEvangelist]

Summary

Phase 1 of CI wall-time optimizations — all changes are low-risk with no behavioural trade-offs. Each item is independently revertable. Background research and tradeoffs are documented in the commit message.

What's in this PR

Frontend / Vitest

• Consolidate test:unit — was 10 sequential vitest run calls, each paying full Vite + TS-transform cold-start cost. Now a single invocation that picks up the same files via vitest.config.ts's include pattern. Individual test:unit:* scripts kept for local granular runs.
• Vitest pool: 'threads' — worker_threads has lower spawn overhead than the default forks pool on Linux. Tests are environment: 'node' with no native addons or thread-unsafe APIs.
• Cache node_modules/.astro keyed on content + astro.config.mjs so Astro's content layer doesn't re-hydrate from scratch when MDX hasn't changed.
• pnpm install --prefer-offline — skips registry HEAD probes when the setup-node pnpm-store cache hits. Falls back to network on misses.
• compression-level: 0 on Playwright artifact uploads — reports contain PNGs, webm videos, and traces that are already compressed.

.NET

• DOTNET_NOLOGO=1 and DOTNET_CLI_TELEMETRY_OPTOUT=1 on both apphost-build.yml and tools-tests.yml.
• tools-tests.yml explicit build + --no-build — dotnet test was implicitly rebuilding each project. Now we restore once, build once, and run dotnet test --no-build --no-restore.

GitHub Actions infra

• concurrency group on ci.yml with cancel-in-progress: ${{ github.event_name == 'pull_request' }}. Stacked PR runs cancel after rebases; pushes to main/release/* are never cancelled.
• changes job uses fetch-depth: 1 plus a targeted git fetch --depth=1 of the PR base and head SHAs. frontend-build.yml keeps fetch-depth: 0 because Lunaria still requires full history.

Verified locally

• pnpm test:unit runs all 159 unit tests across 17 files in 5.5 s with the new consolidated invocation + threads pool.
• YAML for all four workflows parses cleanly.

Explicitly NOT recommended (researched and verified)

Listed here so reviewers don't ask:

• NODE_OPTIONS=--max-old-space-size=* — Node 22+ auto-sizes heap to ~50% of RAM; manual flags are cargo cult on 16 GB runners.
• pnpm --reporter=silent — already silent in non-TTY CI.
• DOTNET_SKIP_FIRST_TIME_EXPERIENCE — deprecated since .NET Core 3.0.
• build:skip-search — Pagefind IS exercised by site-search.spec.ts and ts-api-search.spec.ts.
• Vitest isolate: false — api-markdown.vitest.test.ts uses vi.mock; state-bleed risk isn't worth the marginal win.
• Caching ~/.cache/ms-playwright — Playwright's own docs advise against it.

Follow-ups (not in this PR)

Two higher-reward changes that trade off discipline or runner minutes; I'll open separate PRs once we agree:

• NuGet caching with packages.lock.json — adds dependency-bump discipline.
• Shard Playwright across viewport projects (desktop-chromium, tablet-chromium, mobile-chromium) — 3× runner minutes for the E2E phase in exchange for ~60% wall-time reduction.

[Copilot]

Pull request overview

This PR optimizes CI wall time by reducing redundant test startup/build work and adding workflow-level caching and runner settings without changing product behavior.

Changes:

• Consolidates frontend unit tests into one Vitest invocation and switches Vitest to the threads pool.
• Adds frontend dependency/cache/artifact upload optimizations.
• Updates .NET workflows and CI orchestration to reduce repeated work and cancel superseded PR runs.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/frontend/vitest.config.ts	Sets Vitest to use the threads worker pool.
src/frontend/package.json	Consolidates test:unit into one Vitest run.
.github/workflows/tools-tests.yml	Adds .NET CLI env flags and separates build from test with --no-build.
.github/workflows/frontend-build.yml	Adds pnpm --prefer-offline, Astro content cache, and uncompressed artifact uploads.
.github/workflows/ci.yml	Adds workflow concurrency and shallow PR diff fetches.
.github/workflows/apphost-build.yml	Adds .NET CLI env flags for AppHost build.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[eerhardt]

Hmm, why isn't this the default behavior? I don't understand this change.

[IEvangelist]

Good question. Two things going on:

Why GHA doesn't do this by default: concurrency control isn't universally safe — release pipelines, audit jobs, and workflows with external side-effects need every commit to run to completion. So GHA's default is "run everything in parallel" and concurrency: is the opt-in switch.

What this specific block does:

• PRs — key includes github.event.pull_request.number + cancel-in-progress: true. New commits to a PR cancel the older in-flight run for that same PR. Saves runner-minutes on rapid fixup / rebase pushes.
• Pushes (main / release/*) — key falls through to github.run_id, which is unique per run, so each push effectively gets its own group. Combined with cancel-in-progress: false, push runs neither cancel each other nor queue against each other. Every commit to main always gets its own CI run.

The reason this is split rather than just using a single shared group with cancel-in-progress: false is the bug the bot caught above: when runs share a group, GHA cancels older pending runs regardless of cancel-in-progress. Giving push events a unique group sidesteps that entirely.

[eerhardt]

New commits to a PR cancel the older in-flight run for that same PR. Saves runner-minutes on rapid fixup / rebase pushes.

I thought that is the normal "default" behavior...