Replace num_cpus with std::thread::available_parallelism.

[partim]

This PR removes the dependency on the num_cpus crate by using the standard library’s std::thread::available_parallelism instead.

Fixes #902.

[ximon18]

LGTM, the fns called by both the old and the new code seem to query for some number of "logical" CPUs so that's consistent. One minor point: I did wonder whether it would make sense to extract this repeated code fragment into a helper fn:

available_parallelism().map(|x| x.get()).unwrap_or(1)

[partim]

Sounds like a good idea, indeed. Done!