Profile allocations in benchmarks

[nvzqz]

This adds AllocProfiler, an allocator implementation that tracks the number of allocations and bytes allocated during benchmarks. It can wrap any GlobalAlloc, such as mimalloc.

Example benchmark output:

[nvzqz]

I believe I've pored over this code enough to be convinced that it's correct, so I'm going to merge it. Also, I can't come up with ways to make it faster without spending much more time.