Interpreter P/Invoke support

[kg]

• Adds a custom opcode for pinvoke calls that don't rely on a marshaling wrapper
• Adds a custom C++ library with some functions to call via pinvoke for testing
• Adds a couple basic tests for p/invoke, with more commented out that don't work yet
• Flips the meaning of the helper tag bit (this tag bit will be removed in a followup PR)

[kg]

cc @BrzVlad @janvorli
Main remaining problem here is the check inside the ndirect import worker. I don't really know what the right fix is, other than updating it to be happy seeing an interpreter frame.

[janvorli]

@AaronRobinsonMSFT all the things you've commented on are from my WIP PR that I am still cleaning up, Katelyn has built her stuff on top of that.

[kg]

The working directory change is controversial, so I'm open to changing it around and making InterpreterTester be the one that builds the dll instead.

IMO Interpreter.dll should be built alongside pinvoke.dll since that's the actual dependency, and we should be running it where-it-was-built instead of in InterpreterTester's artifacts dir. IIRC msbuild has never been able to express 'when foo.csproj is referenced by bar.csproj, copy these satellite files along with foo.dll'.

I like being able to directly run Interpreter.dll with corerun in the debugger and have it work correctly without having to know to set a special cwd. But I can see the argument for doing it the other way. This will fail in a less confusing way once we have EH, at least - at that point you would see the dll not found error and be able to figure out what's wrong pretty quickly.

[janvorli]

@AaronRobinsonMSFT do you have any comments? I'd like to get this change in soon.

[dotnet-policy-service]

Tagging subscribers to this area: @BrzVlad, @janvorli, @kg
See info in area-owners.md if you want to be subscribed.

[kg]

Everything works now as long as the prestub for the pinvoke doesn't throw (due to a missing entry point). If the pinvoke throws, our managed catch block runs (in the interpreter, I think) and then the prestub's catch block runs and asserts because there's no exception. It seems like we're probably missing an InlinedCallFrame for the pinvoke like jan mentioned in a chat, which suggests we need a dedicated call opcode for pinvokes even though a regular call mostly works.

[janvorli]

LGTM, thank you!

[janvorli]

/ba-g the Android build timeouts in all recent PRs.