Eliminate circe-yaml dependency

[hubertp]

Pull Request Description

This change adds our very-own YAML parser on top of SnakeYAML. Compared to Circe parser on top of SnakeYAML. The advantage? In some not-so-distant future we might actually get rid of circe and the related performance issues.

The logic is similar to what circe does i.e. analyzing SnakeYAML to build our own structure.
All configs already parseable. We could auto-generate some of the code but still some of the logic would have to be tweaked by hand; the current logic has a number of special cases, as I found out the hard way.

Closes #9113.

Important Notes

It's a bit hard to get a definite number of how things improved but here are some screenshots:
Before:

After:

There are a couple more savings in the range of 10-20ms each which all adds up easily to about 300ms.

Checklist

Please ensure that the following checklist has been satisfied before submitting the PR:

• The documentation has been updated, if necessary.
• Screenshots/screencasts have been attached, if there are any visual changes. For interactive or animated visual changes, a screencast is preferred.
• All code follows the
  Scala,
  Java,
  TypeScript,
  and
  Rust
  style guides. In case you are using a language not listed above, follow the Rust style guide.
• Unit tests have been written where possible.

[hubertp]

In the current state of PR there is zero presence of io.circe.yaml class loading (and usage) during startup despite it being still on the classpath. This is because I'm having trouble with getting rid of the encoding part which is obviously not present during startup. I think we can deal with it in the follow up PR.

[Akirathan]

I would prefer more explicit encoding/decoding. The best would be if the implementation of our SnakeYamlEncoder and Decoder would be in Java. But if this change gives us around 300 ms improvement in startup, that is good enough.

[hubertp]

I would prefer more explicit encoding/decoding. The best would be if the implementation of our SnakeYamlEncoder and Decoder would be in Java. But if this change gives us around 300 ms improvement in startup, that is good enough.

We are using SnakeYAML which is as close to Java as one can get. I understand the sentiment but YAML parsing is still used 99% of the time in Scala code. And implicits eliminate a lot of useless boilerplate code in this case.

[JaroslavTulach]

Overall it looks like a good move forward.