Add CI job integration-chutney-shadow

Refactor chutney test to support being run natively or under shadow, and add a CI test that runs it under shadow.

Unlike when run natively, the test appears to be reliable when run under shadow. It's also a fair bit faster.

I've locally verifies that the native chutney test still (usually) works, but since it's still flaky when run natively, it's still disabled.