Better auto-detection of physical cpus

[fluffynuts]

HT is a lie which just causes things to take longer.

When auto-scaling for the system, we should be trying to use actual cpu count as a limiter. MAX_CONCURRENCY is a way for the user to override this, but we have to be smarter than os.cpus() because that reports logical processors, which essentially counts total HT threads.

here's some ideas:

• windows: https://itsiti.com/check-hyper-threading-enabled-windows/
• linux: https://unix.stackexchange.com/questions/33450/checking-if-hyperthreading-is-enabled-or-not
• OSX: ??