We have exactly one custom runner, so the emphasis that the old
structure put on custom runners was misleading. It's better to
foreground the normal case.

Copy-pasted from `ListBufferBenchmark`, but based on the specifics of
the code changes in this PR, I added benchmarks for `addAll` and
`reverseIterator`, and modified the `insertAll` benchmark so it's
measuring `ArrayBuffer`-into-`ArrayBuffer` insertion (rather than
`Seq`-into-`ArrayBuffer`).

Make `ArrayBuffer`'s iterator fail-fast when the buffer is
mutated after the iterator's creation.