• There is a measurable (on benchmarks) overhead from "auto" priority implementation, which increases prio inside transaction.

The "auto" implementation shuffles low-priority (outside of transaction) clients into work queue, and handles high prio clients(inside transaction), directly, on socket IO notification. The "shuffle" costs some CPU overhead, and can result in more threads than necessary.

So, while we still allow "auto", we rather make it non-default here, because of non-zero overhead.

• As reported by MS folks, setting max.threads too high may have detrimental
  effects on native threadpool, so use native default.