Close netcdf dataset after getting its size

[ninsbl]

Thanks for this cool python library. I looked at alternatives (like siphon) but this one still is the most to the point solution with the features / functions I need.
Hopefully, the fact that there have not been any commits to the master branch the last years is a sign of the reliability of the library and that it just works (and not that it is no longer actively maintained).

When I used it to crawl a larger Thredds server, I noticed that the server at some point returned a 502 Bad Gateway error. It may be related to the issue I try to address in this PR, that netcdf files are not closed after their size is computed, leaving the server with plenty of open datasets?

Another related question is, if I am not interested in the size of a dataset, but just want to get the URLs opening the dataset and computing it`s size is time spent unneccessary. Would it be acceptable for you to change the default, that the size is only computed on user request? I could have a look at that and make a separate PR...

But if this library is no longer maintained, I would be really happy if you could point me to an alternative library that could be used as a replacement (with the same features)...

[ninsbl]

So it seemed the main problem was actually the TCP connection that was kept alive (for some time) with a newer version of requests. That accumulates and led to thousands of open connections on a bigger thredds server, effectively killing the server.

This could probably be solved more elegant, like e.g. here:
https://stackoverflow.com/questions/54876452/run-parallel-request-session-in-python
but closing the connection after the data is read limits the number of open connections to the number of workers. Performance seems unchanged.