Switch CodSpeed benchmarks to instrumented mode

[MatthewMckee4]

Summary

Switches the CodSpeed benchmark job from walltime to instrumented mode. Walltime was running on ubuntu-latest (with a # should be codspeed-macro note) because we can't use CodSpeed's managed macro runners, and shared GitHub VMs are too noisy for reliable walltime numbers. Instrumented mode counts CPU instructions via valgrind, so results are deterministic on any runner.

The bench target (codspeed-divan-compat) supports both modes — the only build-time switch was -m walltime, which is now dropped. Removed the mode: walltime input from the action and the now-unused codspeed-macro label from actionlint.yaml.

Test Plan

ci

[codspeed-hq]

Merging this PR will create unknown performance changes

🆕 1 new benchmark
⏩ 1 skipped benchmark¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
🆕	Simulation	karva_benchmark	N/A	46.1 ms	N/A

---

_{Comparing switch-codspeed-instrumented (4376e52) with main (faef1bf)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩