Allow the NativeAOT runtime pack to be specified as the ILC runtime package

[jkoritzinsky]

Allow specifying the NativeAOT runtime pack as the "runtime package" for ILC to enable us to exclusively ship the NativeAOT runtime through the NativeAOT runtime pack packages without further breaking the in-box dotnet build /t:Publish scenario any more than it already is.

This change also helps move us towards less confusion in VMR/Source-Build scenarios. Today, we have to build a "host ILC" package that targets the SDK's RID with only ILC. If this one is accidentally picked up by any testing in the VMR for runtime assets, things fail with a pretty inscrutable error.
In the future, we'll still have to build the "host ILC" package, but if we move the SDK to only use the NativeAOT runtime pack for "target" assets in .NET 10+, we can ensure that the "host ILC" package doesn't have a shape that differs from the actually shipping ILC package.

[dotnet-policy-service]

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

[filipnavara]

Was this tested in the iOS-like scenarios (which use PublishAotUsingRuntimePack)? Does this depend on any SDK changes?

Also, is this supposed to replace dotnet/sdk#37872?

[jkoritzinsky]

This depends on a future SDK change to actually do anything (that change will actually give back the runtime pack package instead of the ILC target package).

This does not replace dotnet/sdk#37872. That is still planned to be merged in the future, after a solution for the _IsPublishing property is decided.

[sbomer]

Is the intention to get rid of IlcUseNativeAOTRuntimePackLayout once the SDK change flows? LGTM, thank you!

[jkoritzinsky]

Is the intention to get rid of IlcUseNativeAOTRuntimePackLayout once the SDK change flows? LGTM, thank you!

Yes, the plan is once dotnet/runtime is consuming an SDK that always has the NativeAOT runtime pack passed through in a way for this code to recognize, we can remove support for the old infrastructure and update the dogfooding docs.

[jkoritzinsky]

/ba-g WBT failures are unrelated