resolving against a style that inherits

I think our current output for resolving against a style that has it's own resolution against a base style produces selectors of the same specificity, making us susceptible to concat order issues.

Start by reproducing the reported case where a style resolves against a base style and inspect the generated selectors and their specificity. Trace the existing style-resolution entry point, then add or update a focused test demonstrating the inheritance case; it is done when the output no longer permits concatenation-order conflicts.

Written by the indexing model from the issue text.