Restructure micro benchmarks

[ludfjig]

I think we should reorientate the benchmarks, we should have a set of tests that we perform on sandboxes of differing sizes , for each sized sandbox we should:
measure the create time
measure the drop time
have guest calls with a varying number of parameters that does not reset the sandbox
have guest calls with a varying number of parameters that does reset the sandbox.
repeat the above where there are 1-n host function calls made by the guest function
measure the cost of snapshotting when modifying varying proportions of the sandbox memory
measure the cost of restoring when modifying varying proportions of the sandbox memory.
measure the cost of calling with the largest parameters that sized sandbox can support.

Originally posted by @simongdavies in #683 (comment)