👷 fix(tools): route HTTP through httpx2 with tenacity retry

[gaborbernat]

The 🚀 CodSpeed job fetches the CSS and JS minify corpora over HTTP the first time each benchmark runs, and the tools/generate_* scripts pull pinned Unicode, PSL, and IANA data the same way. When several CI runs start close together, GitHub and the npm CDNs answer one of those GETs with 429 Too Many Requests, and because the stdlib urllib calls had no retry a single rate-limited response raised straight out of the run and turned it red. 🚦

Every network fetch in tools/ now goes through one httpfetch.fetch_bytes helper built on httpx2 and wrapped in a tenacity retry. The retry stays narrow: it backs off with capped exponential wait on a dropped connection (TransportError) and the transient server statuses (429, 408, 425, the 5xx family), then re-raises at once anything a retry cannot fix, such as a 404 or 403. The corpus bytes are unchanged, so no benchmark baseline shifts; the fetch just survives a transient rate limit instead of failing the job.

httpx2 and tenacity join the test dependency group because the benchmark registry imports the corpus module during collection; both ship pure-Python py3-none-any wheels for every supported interpreter, the free-threaded builds included.

[codspeed-hq]

Merging this PR will improve performance by 9.42%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
✅ 81 untouched benchmarks
⏩ 18 skipped benchmarks¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	test_feature[text-content]	2 ms	1.8 ms	+9.42%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing gaborbernat:ci/bench-fetch-retry (42c3774) with main (fe744a7)}

Footnotes

1. 18 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩