Use single-threaded executor in benchmarks by default

## What problem does this solve or what need does it fill?

As investigated in https://github.com/bevyengine/bevy/pull/9835, the overhead of multithreading can make benchmarks noisy in ways that are irrelevant to what we're actually testing.

## What solution would you like?

Implement this solution executed in the linked PR on our other benchmarks where relevant. Except when we're explicitly testing system parallelism, we probably don't want it.