Use explicit namespace specifiers when promoting/singling fixity declarations

[RyanGlScott]

During promotion, we now give promoted fixity declarations the type namespace, and during singling, we now give singled fixity declarations the type namespace (if it is a singled class declaration) or the data namesapce (for all other singled names). Doing so makes our intentions clearer, and it also future-proofs the code against planned GHC changes discussed in GHC proposal 65, where specifier-less fixity declarations (e.g., infixl 4 %%%) will become an error if they refer to both a term-level and type-level %%% name.

We now also take namespace specifiers into account when promoting fixity declarations. This is because Wrinkle 1: When not to promote/single fixity declarations in Note [singletons-th and fixity declarations] describes a special case where we must avoid promoting fixity declarations for infix names, but this workaround only needs to be used if the fixity declaration lacks an explicit namespace specifier. If it does have an explicit namespace specifier (e.g., infixl 4 data %%%), then we can promote it (e.g., to infixl 4 type %%%) without fear of name clashes.

Fixes #582.

[RyanGlScott]

I was surprised to see that these generated fixity declarations seemingly lacked namespace specifiers. Some investigation reveals that the declarations actually do have namespace specifiers, but due to a GHC pretty-printing bug, the namespaces don't get printed in the -ddump-splices output. Thankfully, the generated code still behaves as expected, but we'll have to live with some counterintuitive golden output until that GHC bug gets fixed upstream.