C front-end: floating-point built-ins are in the symbol table

It should never be the case that no symbol for `__CPROVER_isgreaterf`,
`__CPROVER_isgreaterd` and similar floating-point built-ins is present
for these are declared in cprover_builtin_headers.h. This commit removes
dead code and adds an additional consistency check.

Author: tautschnig

src/goto-programs/builtin_functions.cpp

@@ -24,7 +24,7 @@ Author: Daniel Kroening, kroening@kroening.com
 #include <util/rational.h>
 #include <util/rational_tools.h>
 #include <util/simplify_expr.h>
-#include <util/symbol_table_base.h>
+#include <util/symbol.h>
 
 #include <langapi/language_util.h>
 
@@ -1473,21 +1473,17 @@ void goto_convertt::do_function_call_symbol(
     // append d or f for double/float
     name+=use_double?'d':'f';
 
+    DATA_INVARIANT(
+      ns.lookup(name).type == f_type,
+      "builtin declaration should match constructed type");
+
     symbol_exprt new_function=function;
     new_function.set_identifier(name);
     new_function.type()=f_type;
 
     code_function_callt function_call(lhs, new_function, new_arguments);
     function_call.add_source_location()=function.source_location();
 
-    if(!symbol_table.has_symbol(name))
-    {
-      symbolt new_symbol{name, f_type, mode};
-      new_symbol.base_name=name;
-      new_symbol.location=function.source_location();
-      symbol_table.add(new_symbol);
-    }
-
     copy(function_call, FUNCTION_CALL, dest);
   }
   else if(identifier == "alloca" || identifier == "__builtin_alloca")