add connection pooling for remote objects

Is your feature request related to a problem? Please describe.
I have a workflow for downloading the latest version of RefSeq. However, there are a few thousand protein files to download. Using FTP remote to download the files is untenable. I gave up after waiting a day for the DAG to resolve.

Describe the solution you'd like
I hacked remote/FTP.py to use connection pooling (https://github.com/niciqy/ConnectionPool) and it was able to resolve the DAG in under half an hour.

For a different workflow with only a dozen files, it cut the dry-run time from 30s to 20s.

I've only looked at the FTP module so far, and I only partially understand how the whole Remote system is built, but it could be a way to speed up any workflow using remote resources.

Describe alternatives you've considered
I can make my workflow functional by manually checking the dates and downloading with wget or curl, but that requires more code in the workflow than simply using remotes.

Start with remote/FTP.py and trace how the broader Remote system creates and reuses connections. Compare the existing FTP behavior with the reported connection-pooling approach, then verify that workflows using many remote resources resolve substantially faster without breaking smaller remote workflows.