[Copilot]

Summary

• Build the Bazel workspace on native Windows for ready pull requests and pushes, using the shared heavy-work gate, covering cfg(windows) code.
• Keep the full Windows test suite opt-in via workflow_dispatch with run_os=true.

Motivation

The Linux Cargo check was the only automatic Windows compilation path, outside the workspace's Bazel build authority.

Changes

• Replace the Clippy job's Cargo check and Cargo-only setup with a native Windows Bazel build:

  bazel build --config=ci //...

• Reuse the existing Windows runner; retain its full Bazel test journey for explicit OS dispatches.

Test plan

• The affected Bazel targets pass
• bazel build //... passes when the change crosses package boundaries
• Tested manually (describe below if applicable)

Checklist

• CHANGELOG.md updated (under [Unreleased] if no version bump)
• No secrets, credentials, or .env files included
• Breaking changes documented (if any)

• Fixes Replace CI's Windows cargo check with a Linux-hosted Windows Bazel check #4324

[changeset-bot]

⚠️ No Changeset found

Latest commit: 740c827

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[ScriptedAlchemy]

Pushed d74895d. Windows Bazel builds now honor the existing run-heavy scope gate, so draft PRs retain the lightweight path. Ready PRs and pushes compile the native Windows workspace; full Windows tests remain available through an explicit run_os dispatch. The workflow YAML parses, and diff/commitlint checks pass. Native Windows compilation remains assigned to CI.

[ScriptedAlchemy]

Pushed ff2342a after the native CI failure reported four agent-hosts journeys timing out at their initial code-index generation.

The failing mount reaches a concrete pre-existing text-build admission defect: it added the watermark margin to the ledger reservation and then shrank it, so a build fitting the actual headroom could be refused by the larger temporary charge. This commit reuses the reviewed #4393 implementation that sizes and charges the build atomically under the canonical ledger, plus the cgroup-headroom regression already reviewed on #4379. It retains the existing ceiling, bounded reclamation, pressure/wakeup rules, and reservation owner.

The journey helper still uses its five-second deadline, 10 ms cadence, and original generation-ready condition. It now captures bounded park, worker, pressure, and ledger evidence inside that same deadline and signals the existing scheduler's cancellation on failure. Reporting after timeout performs no new readiness waits or diagnostic reads.

Validation: the transplanted production files match #4393 byte-for-byte; the added cgroup regression matches #4379. Independent source review, pinned rustfmt, diff checks, and commitlint pass. Native behavior remains for fresh CI. The old timeout log did not include enough admission state to prove this defect caused all four failures, so the diagnostic evidence remains necessary.