Don't measure cost of data allocation in benchmarks

[nolanlawson]

A lot of our benchmark code looks like this:

before(() => {
  // ...
});

run(() => {
  const store = new Store();
  store.runLots();
  // do something
})

Since the store.runLots() (or store.run()) is in the run() hook, we are measuring the cost of allocating a large array of data (in this case, a 10k array). This pointlessly increases the measured time with something that we effectively don't control.

We should move the store-building logic into the before() hook.

[git2gus]

This issue has been linked to a new work item: W-13718118