fix(core): name the ripgrep acquisition stage that failed (TKT-378)

[sepo-eng]

What

Makes the Windows ripgrep failure diagnosable. It does not fix it — the cause is still unconfirmed, and this is deliberate.

TKT-378's diagnosis established that the failure survived an entire investigation uncaused by construction: ripgrep.ts wrapped the whole program — acquiring the binary as well as running it — in a catch-all mapError reporting "ripgrep execution failed", and the reported message discarded the cause the tagged error carried. A stalled download, a shell that never exits, and a genuine execution failure all produced the same string.

Three changes, exactly the ones the diagnosis asked for:

1. The cause is surfaced. Both acquisition and execution failures put the underlying cause in the reported message instead of only in the unread cause field.
2. Acquisition is separated from execution. binary.filepath gets its own label, so a failed download can no longer report itself as a failed execution.
3. Stages are bounded individually. Download (60s) and extraction (30s) each carry their own timeout naming the stage, so a hang says which step hung. Extraction failures also report the interpreter and exit code alongside its output.

Why the test-timeout change is part of it

The four ripgrep suites get a per-test timeout above the acquisition budget. Without it a hung stage still dies on bun's generic 30s test timeout, naming nothing — the labelled failure would never be the one you see, which is the dead end this is meant to remove.

Getting a Windows answer

test.yml has no Windows leg at all (TKT-338 removed it to stop burning GitHub-hosted minutes), so there was no job to re-enable. This adds ripgrep-windows.yml: dispatch-only, running the four ripgrep suites and nothing else. The fork publishes a signed Windows CLI (publish.yml's sign-cli-windows job plus a win-arm64 leg) whose core search fails on first use, and that shipping platform currently has zero automated signal on search. One short run on demand answers it without putting minutes back on every push.

Dispatch it after merge (a workflow_dispatch workflow has to be on the default branch before it can be dispatched), then fix the stage the labelled failure names.

What was verified

• Cold-cache acquisition genuinely runs on Linux. No system rg on PATH, fresh XDG_CACHE_HOME — a 4.5 MB rg appears in the empty cache and all four suites pass. The download→extract→execute chain was exercised, not skipped past a preinstalled binary.
• Each new message was seen to fire, not assumed:
  • forced 404 → ripgrep acquisition failed: ripgrep download failed from <url>: StatusCode: non 2xx status code (404 GET ...)
  • forced download timeout → ripgrep download timed out after <d> from <url>
  • forced extraction timeout → ripgrep extraction timed out after <d> running tar on <archive>
• packages/core full suite: 1216 pass, 0 fail. Typecheck clean (also via the pre-push hook's repo-wide bun typecheck). Prettier and oxlint clean on the changed files.

What is deliberately not here

No PowerShell-quoting fix. Quoting is the most visible difference on the Windows path, not the demonstrated cause, and the evidence — a 30s hang rather than a fast fail — points away from it. Guessing here is what produced the previous dead end.

Linux is refuted on evidence and needs no change; the Windows severity stands.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KbNYZHcnxRCknFmhhPHe7L

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

This PR doesn't fully meet our contributing guidelines and PR template.

What needs to be fixed:

• PR description is missing required template sections. Please use the PR template.

Please edit this PR description to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.