`naga::proc::Namer` uses a different `HashSet`

[kpreid]

#7338, released with naga 25.0.0, changed naga::proc::Namer::reset() to accept a hashbrown::HashSet — but not naga::FastHashSet. This requires callers to have a separate dependency on hashbrown to obtain the type.

Possible fixes:

• Use naga::FastHashSet (breaking change)
• Re-export hashbrown::HashSet

I would presume that there is no reason not to use FastHashSet here, but perhaps there’s a nuance I missed.

[kpreid]

Idea for a bigger-picture solution: naga could provide its own opaque types for "keyword set" and "lazy initialized keyword set", which wrap up the details of which HashSet is used (or if there even is one) and the lazy initialization, so the public usage just looks like "please put one of these in a static and .get_or_init(&[...keywords...]) it".

(Though I still think it would be worth trying storing the keywords list as a simple sorted slice for binary searching, which would make the lazy-initialization part moot.)