This got a bit delayed with Christmas, and then I fell ill, but it is here at last.

The plugin system itself is fairly simple, it essentially consists of a few definitions in a new conda.plugins module.
Initially, I planned to only expose the plugin manager instance to the rest of the conda codebase via the context object, but that proved insufficient right on the first plugin implementation, as the argument parsing happens before the context object is created, and the context object creation is currently inherently dependent on the argument parsing. As such, the plugin manager is also exposed via a helper function in the context module. This might limit us in the future, but that could potentially be solved by refactoring the context object a bit.
Instead of exposing the plugin manager directly, I think we might benefit of having a wrapper around it that does all the data necessary gathering, but this is not clear to me at this point, hopefully it will become clearer as we implement more plugins and get an idea of their needs.

Overall, I think the code size is pretty small, which was my goal. It demonstrates how easily plugins can be implemented in the conda codebase.

Regarding the actual subcommand plugin implementation, I did explore improving things a little bit more in regards to the argparse API usage and how to plug this functionality into it, but ultimately settled on just setting epilog instead of overwriting print_help. Part of the reasoning was that jezdez expressed the wish to move away from argparse to a different module, like click, so spending more time trying to improve this design did not seem worth it. For the same reason, and backwards compatibility, I decided to give plugin users full control over the argument parsing, instead of locking them into a specific API, like integrating the hook directly into argparse

I still need to add the tests, but I think I am reasonably happy with the end result so far.