perf(test-benchmark): consider using pypy to fill benchmark releases

Just a reminder to try this out. In the just benchmark test recipes, I noticed that `pypy` was notably faster than cpython.

Consensus tests didn't show an improvement with pypy (related issue #2029), but it's worth checking whether that's the case with the benchmark tests.

Perhaps it could be beneficial with benchmarks because we're simply doing much more work that can be re-used by the JIT compilation? So JIT startup is better amortized and savings outweigh the JIT compilation?

It remains to be seen if the increased memory bottleneck prevents us from using pypy.