Construct a reliable environment for measuring Hyperlight performance

Because we are currently running Hyperlight with nested virtualization on public clouds, there is a lot of noise in micro-benchmarks, and it is not clear that they are at all useful (e.g. time of day seems to influence some quite a lot). We should find a more reliable environment for running benchmarks, ideally including a dedicated hardware host without NV overhead, isolated cores, frequency pinning, etc. We should separately also still run benchmarks in an NV partition in order to better reflect our expected use cases, but try our best to make this reproducible as well.