.github/workflows: Migrate workflows to Blacksmith runners

[blacksmith-sh]

This PR has been automatically generated by a team member in your GitHub organization using Blacksmith's Migration Wizard. This PR changes the following:

1. Your selected workflows will now run on Blacksmith's 2x faster hardware (e.g., runs-on: blacksmith-4vcpu-ubuntu-2204). Learn more about the different instances available to choose from.
2. Your jobs running on Blacksmith will now have all official GitHub and popular third-party cache actions automatically interact with our 4x faster, colocated cache. Learn more about Blacksmith's actions cache.
3. Your GitHub Actions will now actually be observable. Learn more about Blacksmith's logging and other observability features.
4. Your selected checkout steps now use Blacksmith's cached checkout action. Learn more about Git checkout caching.

[github-code-quality]

Code Coverage Overview

Languages: Rust

Rust / code-coverage/cargo-llvm-cov

The overall line coverage in commit bc7cfe4 in the blacksmith-migration... branch remains at 84%, unchanged from commit 37cc911 in the main branch.

Show a line coverage summary of the most impacted files.

File	main 37cc911	blacksmith-migration... bc7cfe4	+/-
actix-http/src/h1/payload.rs	99%	93%	-6%
actix-http/src/...1/dispatcher.rs	85%	85%	0%
actix-http/src/h2/service.rs	60%	61%	+1%
actix-web-codegen/src/route.rs	72%	73%	+1%
actix-http/src/h2/mod.rs	88%	91%	+3%
actix-web/src/t...est_services.rs	52%	57%	+5%
actix-multipart...rc/extractor.rs	42%	56%	+14%

---

_{Updated September 30, 2026 17:45 UTC}

[robjtede]

All four full reruns passed on bc7cfe4. Compared with warm main CI at the exact base commit 758f6ee, CI was 11% quicker end to end: 15m41s → 13m56s, saving 1m45s including queue time. The first successful full run on this head took 20m43s with mixed caches; the warm rerun was 33% shorter. Main CI #5786, initial successful CI, attempt 1, warm CI, attempt 2.

Job times below exclude queueing. On the initial successful run, Linux and Windows missed their caches; macOS and doc tests already had full cache hits. All six matrix jobs and doc tests reported full cache hits on the rerun.

CI job	Main at base, warm	Main median, 5 runs	Initial successful PR run	Warm rerun	Warm vs base
Linux / msrv	4m23s	4m40s	9m44s	8m04s	84% longer
Linux / stable	7m31s	5m39s	10m50s	6m44s	10% shorter
macOS / msrv	11m30s	10m34s	2m43s	2m48s	76% shorter
macOS / stable	12m31s	12m31s	2m22s	2m31s	80% shorter
Windows / msrv	15m20s	13m18s	18m15s	12m32s	18% shorter
Windows / stable	12m32s	13m21s	13m23s	11m06s	11% shorter
doc tests	2m56s	1m20s	1m13s	1m20s	55% shorter
Read MSRV	0m15s	0m14s	0m13s	0m13s	13% shorter

Summed execution time across all eight CI jobs fell 32%, from 66m58s on main to 45m18s on the warm rerun (initial PR run: 58m43s). macOS jobs were about 4–5× as fast as the exact-base sample. Linux / MSRV was 84% longer, despite a full cache hit.

The five recent successful main-push CI runs ranged from 14m07s to 19m24s, with a 17m44s median. The warm rerun was 21% shorter than that median. Samples: #5786, #5777, #5760, #5744, #5739. The job medians show that Linux / stable was also slower than its recent main median.

Other full reruns:

Workflow / job	Old-configuration PR sample	Initial run on this head	Warm rerun
Lint wall time, including queueing	0m33s	1m07s	0m52s
Clippy execution	0m28s	0m54s	0m40s
Semver wall time, including queueing	1m40s	6m07s	1m29s
Semver execution	1m37s	5m57s	1m18s
Coverage wall time, including queueing	—	7m35s	9m22s
Coverage execution	—	6m42s	9m09s

Clippy and Coverage were already warm on the initial run; Semver was cold. All three reported full cache hits on the rerun. Clippy showed no speedup against the old-runner sample, and Coverage was slower on its rerun. Lint's summed job time was 1m52s / 2m39s / 2m16s, respectively; the disabled external-types job is excluded.

Lint and Semver have no main-push run, so their old-configuration baselines are successful PR runs at 85c20a5, not main CI. The exact-base main Coverage run failed, so it is excluded from speedup claims. Old Lint, PR Lint, old Semver, PR Semver, PR Coverage, failed main Coverage.

Earlier startup failures and the superseded, cancelled CI run provide no complete cold-run baseline and are excluded. These are observed workflow timings from one full rerun, including the runner and setup changes.

[Sruhvx-jpg]

dammm blacksmith really replied to u really fast