Memory usage tests

[nolanlawson]

It would be convenient if Best could measure not only CPU time, but also memory usage. I can envision two types of tests here:

1. Memory leaks – e.g. take a snapshot, repeat some action n times, take another snapshot, diff the two
2. Absolute memory usage – e.g. take a snapshot at a point in time (maybe after a GC+delay) and report it

[nolanlawson]

Details:

1. For the leak scenario, I envision taking two heapsnapshot files and comparing them, because that's the most convenient way to figure out what actually leaked
2. Fo the absolute memory usage scenario, I envision just adding the Chrome/Puppeteer flag that allows precise memory stats and then use window.performance.memory.

[nolanlawson]

Just talked with @diervo, conclusions:

• adding the ability to automatically run 1 iteration of each benchmark and measure the memory before/after (using special Chrome flag to expose precise memory) is reasonable
• users can create their own leak tests by writing a test where it allocates a bunch of stuff and then de-allocates, so it would expect the diff to be ~0
• putting chrome in precise memory mode should not affect CPU perf
• memoryLeakIterations could default to 1
• might call it a "measure"