dotnet trace collect-linux usability issues

During investigations we've noticed a few usability issues.

- [x] When the user provides a host namespace PID to trace a process running inside a container, dotnet-trace can't resolve the diagnostic server socket created by the app in the container. This is tracked by https://github.com/dotnet/diagnostics/issues/5694

- [x] When the user provides a host namespace PID to trace a process running inside a container, the library that backs collection fails to see the process as a .NET process and subsequently all events that are supposed to be registered and emitted through user_events never come through. Root cause has been identified and is tracked in https://github.com/microsoft/one-collect/issues/234

- [x] Requesting CPU or thread time traces results in broken stacks, showing shallow stacks instead of the full expected stacks. This is tracked in https://github.com/microsoft/one-collect/issues/235

We should also document the expected PID to provide and when this might not work as expected.