sm6.h is only used by dxil.c. Perhaps that will change in future commits, but until it does, its contents simply belong in dxil.c.
There's a good chance we'll want to emit sm6 from the HLSL compiler. Why not be proactive, and avoid the code churn?
The practical answer to that question is "to make the life of your reviewers easier and have your patches go in smoothly". I.e., you're essentially asking a reviewer to take a guess at potential future uses for these, or look at later patches in the series. This could be mitigated by writing a good commit message, of course, but even in that case you may end up in a discussion about future usage which would otherwise not be a concern for the patch in question.
To that point, it's not clear to me that we'd want a sm6.h for outputting sm6 from the HLSL compiler either; I think the code for that should just go into dxil.c as well.