fix: resolve dependent generic defaults during type inference

[raushan728]

This PR fixes the FIXME in hir-ty where dependent generic defaults (like struct T<A, B = A>) result in {unknown} when using a placeholder _.

The fix re-applies defaults after insert_type_vars in path.rs to ensure nested inference variables are correctly unified.

[ShoyuVanilla]

That test is just sensitive to query graph changes, not flaky. You could just update the expectation to the actual output

[raushan728]

thanks for the correction @ShoyuVanilla . I've updated the test expectations. I realized my previous local run was mistaken—the test is indeed sensitive to the query changes from this fix.

[ChayimFriedman2]

I'm really not pleased with the solution here, and given that we need to get rid of the distinction between TyLoweringContext and InferenceContext anyway (as I said on Zulip), I would much prefer to solve this by that. That is larger work, of course.

[raushan728]

Thanks @ChayimFriedman2. I just saw your messages on Zulip, let's discuss it further there."