test: fix flaky template download and build lookup failures

[l2ysho]

Note

TL;DR Fixes the 3 open apify-cli flakes from the flake triage log (2026-09-24 sweep). Two of them, plus one extra sighting, share one cause: the template download failed, and apify create exited early without a visible error.

Changes

• Retry transient template fetches (src/lib/utils.ts, src/commands/create.ts): the manifest fetch retries on any error. The archive download retries on network errors and HTTP 408/429/5xx. 3 attempts, 1s then 2s delay. Other 4xx errors still fail at once.
  • Fixes python-scrapy-template-works.test.ts (4/48) and python/works-with-spaces-in-path-to-actor.test.ts (2/48). Also fixes a single ENOENT sighting in propagates-non-zero-exit-code.test.ts.
  • Evidence: in every failing job, the whole test file took 265–770ms. A normal run takes 12s or more, because create installs dependencies. The ENOENT paths (my_actor/main.py, src/) show the archive was never unzipped. Vitest runs silent, so the create error never reached the log.
• Wait for the latest build (test/__setup__/build-utils.ts): call.test.ts and task/run.test.ts now poll taggedBuilds.latest (60s limit) and do not read builds().list() right after push. That list can still be empty, which caused lastBuild to be undefined in call.test.ts (2/48).
• Unit test for the download retry.

Follow-ups

• push.ts, templates/ls.ts and wrapScrapyProject.ts also call fetchManifest() without a retry. I left them out to keep this diff small.

🤖 Generated with Claude Code