Split the processing of workflow-based and chat-based runs

[thedadams]

If too many workflow-based runs are running, then this can lead to chat-based runs not getting processed. We should have a way to split the processing of runs so that this backup does not occur.

[thedadams]

To test this:

1. Run Obot with NAH_THREADINESS=1.
2. Save the following tool (or an equivalent version) to your local machine:

Name: sleep
Description: Sleep for some number of seconds
Args: seconds: The number of seconds to sleep

#!/usr/bin/env sh
sleep ${SECONDS}

3. Add this tool in the UI.
4. Create a workflow.
5. Add a step to this workflow, adding the above tool and giving it the following instructions: Sleep for 120 seconds
6. Run this workflow from the UI. Optionally, in another browser tab, run the workflow again. You should see that the second workflow's run doesn't start until the first one finishes.
7. In another browser window, chat with any agent. This should succeed without waiting for the workflows to finish.