Track Hyperlight microbenchmarks

We should track the performance of Hyperlight on microbenchmarks in a somewhat-reliable environment, so that we understand where the time is actually going and where we can improve it.

Things that we should track include at least:
- RTT for a guest call, with various sizes of parameters
- Time to map/unmap various kinds of host memory regions
- Time to create, snapshot, restore, and drop a sandbox
- All of the above in a multithreaded use case, with various degrees of parallelism (including both under- and over-subscribed compared to the number of cores)