Improve output logging

[rabernat]

Right now, logging is a mix of the original structlog stuff written by @yuvipanda and some click.echo statements I threw in. This is a bit of a mess. Someone needs to dig into this and figure out:

• What we need to log
• What's the best way to do it
• How to integrate click.echo with traditional logging

Since binderbot is designed to be used in CI, good and pretty logging is very important.

[yuvipanda]

I'm thinking about using https://github.com/jupyter/telemetry/ for logging here. It can provide structured, documented logs that can be easily read by other downstream projects if needed. How does that sound? If it feels too heavyweight, we can just use regular structured logging...

[rabernat]

Sounds great to me Yuvi. I don't have much expertise here, so I will happily defer to other experts on what is the best way to do it.

[yuvipanda]

I spent some more time in the binderbot codebase, and have come to different conclusions!

hubtraf's output was primarily for machines, so extreme structure was important. Binderbot's is primarily for humans, so human readability is much more important.
For the most part, detailed timing info or explicit messages about what was currently happening aren't as important as status messages. We should have a debug flag that enables verbose logging, but provide user-readable messages as the primary mechanism.

As such, I think we should remove all non-debug logging anywhere other than from the CLI, and structure our application well enough that we can produce enough useful human output from the cli code.