Add NativeAOT test

[adamsitnik]

There was one warning, about the usage of Array.CreateInstance. IMO we can't get rid of it until we implement #1638

I've disabled this particular warning just to make sure we don't add any new warnings in the meantime. It's not perfect, but still better than no tests at all.

[adamsitnik]

@jonsequitur the Ubuntu leg is most likely failing due to lack of native dependencies:

https://github.com/dotnet/runtime/blob/main/src/coreclr/nativeaot/docs/prerequisites.md

Should we add these dependencies or disable the test for this OS?

[adamsitnik]

@jonsequitur there are other issues that we need to consider:

1. contributors who don't have the https://github.com/dotnet/runtime/blob/main/src/coreclr/nativeaot/docs/prerequisites.md installed are going to get test failures
2. it takes a lot of time to build an AOT app
3. the test uses latest version of NativeAOT compiler. If there is a new bug introduced to it, the CI might suddenly start failing. It happened to us in BDN and we actually have two kind of tests: using hardcoded version to ensure to regressions in BDN and using latest to ensure it's supported

[jonsequitur]

@jonsequitur the Ubuntu leg is most likely failing due to lack of native dependencies:

https://github.com/dotnet/runtime/blob/main/src/coreclr/nativeaot/docs/prerequisites.md

Should we add these dependencies or disable the test for this OS?

I'd prefer to add the dependencies. I'll look into it.

[jonsequitur]

@jonsequitur there are other issues that we need to consider:

1. contributors who don't have the https://github.com/dotnet/runtime/blob/main/src/coreclr/nativeaot/docs/prerequisites.md installed are going to get test failures
2. it takes a lot of time to build an AOT app

We've been using ReleaseBuildOnlyFact and ReleaseBuildOnlyTheory for tests that are important to run during CI but would cause friction during the dev loop, such as complex setup or tests that are necessarily slow.

3. the test uses latest version of NativeAOT compiler. If there is a new bug introduced to it, the CI might suddenly start failing. It happened to us in BDN and we actually have two kind of tests: using hardcoded version to ensure to regressions in BDN and using latest to ensure it's supported

My thought is the hardcoded version to prevent regression is simpler and more important, since here we're testing System.CommandLine and not the AOT compiler. Ingesting new dependencies dynamically can break this CI pipeline, so I would tend to avoid that.

[jonsequitur]

@adamsitnik The update to Ubuntu 20.04 is merged. Could you rebase this and see if the tests now pass?

[adamsitnik]

The update to Ubuntu 20.04 is merged. Could you rebase this and see if the tests now pass?

@jonsequitur I've rebased the branch but the test is till failing on Ubuntu. Is this repo using CI machines from MS internal feed? Or some public ones and we can define what is needed?

This is what we needed in BDN:

https://github.com/dotnet/BenchmarkDotNet/blob/ca51035a62ea89de6059c2822c4a5e5df58560a3/azure-pipelines.Ubuntu.yml#L17-L22

[jonsequitur]

Do you want to try including that in your PR?

[adamsitnik]

@jonsequitur it's finally green!