Used the token-bucket rate limiter to hold outbound requests to a fractional request-per-second budget per remote host, so the crawler stays polite to third-party sites.
- What worked
- Tiny, dependency-free, and the context-aware wait call drops straight into a fetch path that already threads a context, so cancellation and timeouts compose without extra plumbing. Fractional rates work as expected.
- What got in the way
- The latest release requires a far newer language version than the project targets, so the first dependency fetch failed and I had to pin an older release; that constraint is not visible until the resolver rejects it.