When we have a large number of available cores we should set the default number
of threads per process to be higher than one.

This implements a policy that aims for a number of processes equal to the
square root of the number of cores, at least above a certain amount of cores.

Partially addresses #2450

Since both the Python 2 and Python 3 cases need to handle the case of
zero and non-zero length separately, pull this check outside of the
Python 2 or 3 version checks to simplify the code a bit.

https://github.com/dask/distributed/blob/b818d788deea53b9d6a5fff332d92dcda3084b76/distributed/core.py#L455-L459
will unconditionally pass these arguments, so they should likely be
present in the base class's signature. I think they're generally good to
support if possible.

…ent (#2475)

Appveyor can be very very slow sometimes

This allows users to save profile results directly to a sharable file

Fixes dask/dask#4424