Initial upload/instantiation of custom nodes

[reelmatt]

We probably want to hold off on merging until we figure some issues out, but this PR has an implementation of adding custom nodes. Other approaches/questions are discussed in #67.

What this does

• Provides a POST /workflow/custom_node endpoint that accepts a file and adds it to a custom_nodes directory within pyworkflow.
• Provides a GET /custom_nodes endpoint that parses any files located within custom_nodes. This returns a JSON-representation of the nodes similar to the GET /nodes endpoint for "system nodes". Adds to the existing GET /nodes endpoint for system nodes, but now adds searching/parsing of files located within custom_nodes.
• For custom nodes containing packages not currently installed/available, it adds a message telling the user to install the packages and restart the server. it compiles a list of packages and is keyed as missing_packages in the JSON response.
• Adds creating custom nodes to the node_factory.

What this does not do/things to fix

• Custom nodes are currently stored in pyworkflow/custom_nodes so that they can import the Node class to extend. We may or may not want to stored custom nodes separately from the package, in which case we'll need a way to reference Node.
  • The custom node path is hardcoded. We should provide a way to dynamically specify a path (within the root_dir, or elsewhere).
• There's a lot of duplicate code to handle the uploading of the custom node file. It mirrors a lot of the upload_file methods in workflow.py and the Workflow view. If we want to pursue this approach, we should refactor uploading to accept different kinds of files and for different purposes.
• This does not install any missing packages. Execution works for packages specified in our Pipfile (like pandas), but this should be addressed.

[reelmatt]

Latest commits don't change much from the initial PR. It does refactor a lot of node-parsing code (with check_missing_packages being a notable duplicated method still).

The latest commits add Custom Nodes to the same GET /nodes endpoint for the node listing. It does this through a second, separate, for loop for files within pyworkflow/custom_nodes/.

[reelmatt]

I'm working on a refactored approach that handles Nodes differently than this PR. Closing for now, but we can revert to this state of implementation if the refactor is too problematic.