This pull request allows us to set relevant context for the current process and also includes a plug that builds up the context for the current request.

The new ErrorTracker.put_context/1 function uses the process dictionary to store the given context. The context is additive which means that we can call ErrorTracker.put_context/1 multiple times and each call will append to the previous context. If we set twice the same key, the last value will override the previous ones.

There is also a new ErrorTracker.Integrations.Plug.RequestContext (which may need a better name) that automatically populates the context with relevant data for the current request such as: method, path, query string, headers, etc.

The Oban integration has been updated to set the context when the job starts instead of only during a captured exception.

I made the choice of using dots to namespace the context keys. For example we have "request.host" and "request.method" instead of "request" => %{"host" => "", "method" => ""}.

I think that having plain keys by default makes them easier to override by clients if they want to. To override an existing key or add a new one to an existing namespace you just add that key and call it a day. If we used the nested map approach you would need to handle deeply nested keys which makes it more difficult.

In the UI we can use the dot as namespace separator and group the keys as we please. I am still not 100% sure about this so I am open to suggestions.