Remove Docker dependency from Maven testing

[lazar-mitrovic]

ATM we are using Docker to create layered images in order to have a reproducible builds for testing the Maven plugin.
See the rationale behind this decision here, and related discussion here.

This is creating an overhead by fetching an entire docker image with a duplicate GraalVM installation, a loads of additional logic related to Gradle <-> Docker interaction, but the biggest problem is that it limits Maven testing to Linux (since guest Docker image will always be Linux based).

I propose that whole Docker related part gets removed and replaced with Groovy / Kotlin logic which would invoke mvnw to download separate maven installation, custom settings.xml and environment variables which would change .m2 location, and a "folder-juggling" mechanism to simulate repository caching between each run.

I'll work on PoC for this and link it to this issue as soon as it's ready.

[melix]

One important aspect of this is to make sure that the locally built artifacts are fetched from a local repository which wouldn't get polluted by Maven. In other words, we should preserve the fact that we have a local repository which contains artifacts produced by this build (the common libraries and the Maven plugin), behaving like a remote repository and not like the local m2 cache.

Then we should have a local m2 cache which is pre-populated, but restorable in a state where those libraries are not found in the cache when a test starts. This should give us relative confidence that the plugin is working correctly, not perfect isolation but not having the Docker overhead.