A continuation of work in PR #349.

@jackparmer @jody @tarzzz Below is a discussion of two issues with the CSV connector that will affect the onprem setup.

Up to now, connections in Falcon are permanent. That is, when a user connects to a database, the connection is stored in ~/.plotly/connector/connections.yaml. If we close Falcon and we open it again, the connections in ~/.plotly/connector/connections.yaml are restored.

This behaviour wouldn't be possible, unless a copy of the uploaded CSV file is stored in the Falcon server. I'm sure we don't want the Falcon server to become an online file storage.

For this reason, I'm planning to work around it, at the start up of Falcon, by removing from ~/.plotly/connector/connections.yaml all the connections corresponding to uploaded CSV files.

Does that sound OK?

An alternative approach would be for the CSV connector to accept file:// URLs. But this would only make sense for the desktop app.

Another issue that I haven't addressed in #349 is that, users can try to upload a CSV file so large that the Falcon server runs out of memory.

I guess we could have a setting that limits the total amount of memory taken by CSV files.

What do you think about having this setting?