Updated internal interface

[UebelAndre]

This change is a followup to #122 (comment)

[UebelAndre]

@ted-xie Sorry about the delay in responding to your feedback. Continuing the thread here:

Okay. This would be an important detail to leave in a comment or doc somewhere, since it's not obvious.

On this topic: Can you think of a better way to have an end-to-end test for this feature? Maybe use rules_bazel_integration_test?

I am not familiar with rules_bazel_integration_test but to hopefully help prevent regressions, I've moved use of the custom attributes to each rule into that rule's implementation. As long as new attributes are handled there or are common then I think the risk of regression is low.