[FEATURE] A smarter way to determine default npartitions in Dask

[goodwanghan]

Is your feature request related to a problem? Please describe.

Currently, the default number of partitions is hardcoded to 16, the number of cores is hardcoded to 2.

Describe the solution you'd like

We can use psutil and multiprocessing to try to get CPU count, if not found, we hard code. It should try psutil first, if the package is not installed, try multiprocessing. psutil is preferred because it can return physical cores. But I am not sure if we should make this a requirement or not.

[goodwanghan]

Dask already has CPU_COUNT see https://github.com/dask/dask/blob/007cbd4e6779c3311af91ba4ecdfe731f23ddf58/dask/system.py#L53

So we will use it directly.