Add optional CPU cycle measurement to graviola-bench

[brian-pane]

• This was motivated by the observation that the elapsed time for the Graviola benchmarks on macOS can sometimes vary by 10% from one run to the next, probably due to some combination of varying CPU frequency and random background daemons. In past projects, I've found that the CPU's cycle counter, while still noisy, is less noisy than the wall clock.
• This change adds a feature __bench_cycles that replaces the standard Criterion elapsed time metric with a measurement of elapsed CPU cycles for the process on aarch64.
• It should be straightforward to add x86_64 support in a later change, either with inline assembly or by using the Linux perf API. (The latter could also provide access to metrics like instructions retired and branch mispredictions.)
• Because the counter-timer register is per-core, this implementation may return incorrect results if a single benchmark runs long enough for the OS to reschedule it to a different core or to interrupt the execution with some other process. It may be beneficial to add CPU pinning in the future.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.78%. Comparing base (9fc602f) to head (f4a45d9).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #210   +/-   ##
=======================================
  Coverage   99.78%   99.78%           
=======================================
  Files         216      216           
  Lines       56507    56507           
  Branches      230      230           
=======================================
  Hits        56384    56384           
  Misses         92       92           
  Partials       31       31           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[codspeed-hq]

Merging this PR will degrade performance by 10.38%

⚠️ 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

❌ 2 regressed benchmarks
✅ 171 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	rustcrypto[32B]	3 µs	3.4 µs	-10.69%
❌	rustcrypto[32B]	3.3 µs	3.6 µs	-10.06%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing brian-pane:cycles-aarch64 (f4a45d9) with main (9fc602f)}

[brian-pane]

I've found an easier way to get repeatable benchmark results on aarch64: on a quiet Linux host, the elapsed times exhibit much less variability than I'm seeing on macOS.