perf: improve specializer cache keys (#11310)

This PR makes the specializer (correctly) share more cache keys across
invocations, causing us to produce less code bloat.

We observed that in functions with lots of specialization, sometimes
cache keys are defeq but not BEq because one has unused let decls
(introduced by specialization) that the other doesn't. This PR resolves
this conflict by erasing unused let decls from specializer cache keys.

Author: hargoniX

src/Lean/Compiler/LCNF/Specialize.lean

@@ -238,6 +238,7 @@ def mkKey (params : Array Param) (decls : Array CodeDecl) (body : LetValue) : Co
   let key := ToExpr.run do
     ToExpr.withParams params do
       ToExpr.mkLambdaM params (← ToExpr.abstractM body)
+  let key ← Meta.MetaM.run' <| Meta.transform (usedLetOnly := true) key
   return normLevelParams key
 
 open Internalize in